//! `query_log` and its `domains` dimension table in `querylog.db`.
//!
//! Two shapes live here. The free functions follow the milestone-4 repository
//! idiom — prepare, use, finalize — because retention runs them a handful of
//! times per day, and the API read layer at the bottom of the file runs once
//! per HTTP request. The flush loop is the one hot path in the program, so it
//! gets `BatchWriter`, which owns its three statements for its whole life
//! (`db.zig:360` names this file as the reason `db.zig` carries no statement
//! cache).
//!
//! Every string in a `Row` is borrowed for the duration of the call only:
//! `Stmt.bindText` binds with `SQLITE_TRANSIENT`, so SQLite copies before
//! `writeBatch` returns.
//!
//! The rows are expendable log data. Nothing here retries, and the caller
//! decides what a failed batch means.

const std = @import("std");
const Allocator = std.mem.Allocator;

const db = @import("../db.zig");
const provenance = @import("../provenance.zig");

/// One `query_log` row. The logger applies the privacy transforms of PLAN
/// §11.4 before it builds this, so every domain-bearing field is already
/// whatever the operator agreed to store.
///
/// A `null` text field is a fact the query did not have — no upstream was
/// attempted, no rule matched, no CNAME was uncloaked — and reaches the column
/// as NULL. The three closed enums have no such state: every logged query has a
/// policy verdict, a reason for it and a route, even when the verdict is
/// "not evaluated".
pub const Row = struct {
    timestamp: i64,
    domain: []const u8,
    client_ip: []const u8,
    qtype: ?u16,
    qclass: u16,
    /// Twelve bits: the EDNS extended RCODE the client saw. The column's
    /// `CHECK` bounds it to the same range, so a value this type cannot hold
    /// is one the schema would have refused anyway.
    rcode: u12,
    blocked: bool,
    response_time_us: ?i64,
    cache_hit: ?bool,
    upstream: ?[]const u8,
    group_id: ?i64,
    group_name: ?[]const u8,
    policy_action: provenance.PolicyAction,
    policy_reason: provenance.PolicyReason,
    matched: ?[]const u8,
    source_id: ?i64,
    source_name: ?[]const u8,
    cname_target: ?[]const u8,
    safe_search_target: ?[]const u8,
    route_kind: provenance.RouteKind,
    forward_zone: ?[]const u8,
};

const insert_domain_sql = "INSERT OR IGNORE INTO domains (domain) VALUES (?1)";

const select_domain_sql = "SELECT id FROM domains WHERE domain = ?1";

const insert_row_sql =
    \\INSERT INTO query_log
    \\  (timestamp, domain_id, client_ip, qtype, blocked,
    \\   response_time_us, cache_hit, upstream, qclass, rcode,
    \\   group_id, group_name, policy_action, policy_reason, matched,
    \\   source_id, source_name, cname_target, safe_search_target,
    \\   route_kind, forward_zone)
    \\VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10,
    \\        ?11, ?12, ?13, ?14, ?15, ?16, ?17, ?18, ?19, ?20, ?21)
;

/// Owns the prepared statements of the flush loop. Init once, reuse per batch.
///
/// `database` must outlive the writer and must not move: every `Stmt` holds a
/// `*Db`. Neither `Db` nor `Stmt` is thread-safe, so one writer belongs to one
/// task.
pub const BatchWriter = struct {
    database: *db.Db,
    insert_domain: db.Stmt,
    select_domain: db.Stmt,
    insert_row: db.Stmt,

    pub fn init(database: *db.Db) db.Error!BatchWriter {
        var insert_domain = try database.prepare(insert_domain_sql);
        errdefer insert_domain.deinit();
        var select_domain = try database.prepare(select_domain_sql);
        errdefer select_domain.deinit();
        const insert_row = try database.prepare(insert_row_sql);
        return .{
            .database = database,
            .insert_domain = insert_domain,
            .select_domain = select_domain,
            .insert_row = insert_row,
        };
    }

    pub fn deinit(self: *BatchWriter) void {
        self.insert_row.deinit();
        self.select_domain.deinit();
        self.insert_domain.deinit();
    }

    /// One transaction for the whole batch. Domains are interned through
    /// `INSERT OR IGNORE` followed by `SELECT id`.
    ///
    /// On any failure the transaction rolls back, so a batch is all or
    /// nothing, and the writer stays usable for the next batch.
    pub fn writeBatch(self: *BatchWriter, rows: []const Row) db.Error!void {
        if (rows.len == 0) return;

        var tx = try db.Tx.begin(self.database);
        // `errdefer`s run in reverse: the statements are released before the
        // ROLLBACK, so no read cursor is still open when it runs.
        errdefer tx.rollback();
        errdefer self.resetAll();

        for (rows) |row| {
            const domain_id = try self.internDomain(row.domain);
            try self.write(row, domain_id);
        }
        try tx.commit();
    }

    fn internDomain(self: *BatchWriter, domain: []const u8) db.Error!i64 {
        try self.insert_domain.reset();
        try self.insert_domain.bindText(1, domain);
        try self.insert_domain.exec();

        try self.select_domain.reset();
        try self.select_domain.bindText(1, domain);
        // The insert above either created the row or found it already there,
        // so a miss means the table changed under this connection.
        if (!try self.select_domain.step()) return error.NotFound;
        const id = self.select_domain.columnInt(0);
        // A statement stopped on a row keeps its cursor open until it is
        // reset; the transaction must not carry that to the next row.
        try self.select_domain.reset();
        return id;
    }

    fn write(self: *BatchWriter, row: Row, domain_id: i64) db.Error!void {
        var stmt = &self.insert_row;
        try stmt.reset();
        try stmt.bindInt(1, row.timestamp);
        try stmt.bindInt(2, domain_id);
        try stmt.bindText(3, row.client_ip);
        try bindIntOrNull(stmt, 4, if (row.qtype) |v| @as(i64, v) else null);
        try stmt.bindBool(5, row.blocked);
        try bindIntOrNull(stmt, 6, row.response_time_us);
        try bindIntOrNull(stmt, 7, if (row.cache_hit) |v| @as(i64, @intFromBool(v)) else null);
        try stmt.bindTextOrNull(8, row.upstream);
        try stmt.bindInt(9, row.qclass);
        try stmt.bindInt(10, row.rcode);
        try bindIntOrNull(stmt, 11, row.group_id);
        try stmt.bindTextOrNull(12, row.group_name);
        try stmt.bindText(13, @tagName(row.policy_action));
        try stmt.bindText(14, @tagName(row.policy_reason));
        try stmt.bindTextOrNull(15, row.matched);
        try bindIntOrNull(stmt, 16, row.source_id);
        try stmt.bindTextOrNull(17, row.source_name);
        try stmt.bindTextOrNull(18, row.cname_target);
        try stmt.bindTextOrNull(19, row.safe_search_target);
        try stmt.bindText(20, @tagName(row.route_kind));
        try stmt.bindTextOrNull(21, row.forward_zone);
        try stmt.exec();
    }

    /// Best effort: this runs on the failure path, where the error that
    /// matters is the one already on its way to the caller.
    fn resetAll(self: *BatchWriter) void {
        self.insert_row.reset() catch {};
        self.select_domain.reset() catch {};
        self.insert_domain.reset() catch {};
    }
};

fn bindIntOrNull(stmt: *db.Stmt, idx: c_int, value: ?i64) db.Error!void {
    if (value) |v| return stmt.bindInt(idx, v);
    return stmt.bindNull(idx);
}

/// What one prune did, and where coverage now begins.
pub const PruneResult = struct {
    deleted: i64,
    /// The watermark after the prune, which is what a later `availableSince`
    /// will return. Handed back so the caller need not re-read it.
    available_since: i64,
};

/// Deletes every `query_log` row strictly older than `cutoff_ts` and advances
/// the coverage watermark to the same cutoff, in one transaction.
///
/// **The two are one operation, not two.** The watermark is the promise that
/// every query since it is still in the file; a delete that commits without the
/// advance breaks that promise, and an advance that commits without the delete
/// hides rows the file still holds. Either failure rolls both back, and the
/// caller retries the whole thing on its next pass.
///
/// The watermark never moves backward: `max` is what makes a prune with a
/// cutoff older than the file's own creation a no-op on it rather than a
/// regression. Orphaned `domains` rows stay — it is a dimension table,
/// re-interning a name costs one indexed insert, and §11.3 asks for no
/// collection.
pub fn pruneOlderThan(database: *db.Db, cutoff_ts: i64) db.Error!PruneResult {
    var tx = try db.Tx.begin(database);
    errdefer tx.rollback();

    var deleting = try database.prepare("DELETE FROM query_log WHERE timestamp < ?1");
    defer deleting.deinit();
    try deleting.bindInt(1, cutoff_ts);
    try deleting.exec();
    const deleted = database.changes();

    var advancing = try database.prepare(
        "UPDATE querylog_meta SET available_since = max(available_since, ?1) WHERE id = 1",
    );
    defer advancing.deinit();
    try advancing.bindInt(1, cutoff_ts);
    try advancing.exec();

    const watermark = try database.queryInt("SELECT available_since FROM querylog_meta WHERE id = 1");
    try tx.commit();
    return .{ .deleted = deleted, .available_since = watermark };
}

/// The oldest timestamp this file can still answer for. A query window that
/// starts before it is incomplete, and the API says so rather than charting the
/// gap as zero.
pub fn availableSince(database: *db.Db) db.Error!i64 {
    return database.queryInt("SELECT available_since FROM querylog_meta WHERE id = 1");
}

/// `PRAGMA wal_checkpoint(TRUNCATE)`: moves the WAL into the database and
/// truncates it to zero bytes, which is what keeps a day of log writes from
/// growing the WAL past the free space the disk monitor watches.
///
/// SQLite reports a checkpoint blocked by a concurrent reader in the row it
/// returns, not as an error code, so a blocked checkpoint is not an error
/// here. Retention checkpoints after every prune, so the next pass retries.
/// On a database that is not in WAL mode the pragma is a no-op.
pub fn checkpointTruncate(database: *db.Db) db.Error!void {
    return database.exec("PRAGMA wal_checkpoint(TRUNCATE);");
}

/// Rewrites the whole file. Retention runs this rarely by design — on an SD
/// card a full rewrite is the most expensive thing this program does.
pub fn vacuum(database: *db.Db) db.Error!void {
    return database.exec("VACUUM;");
}

pub fn countRows(database: *db.Db) db.Error!i64 {
    return database.queryInt("SELECT count(*) FROM query_log");
}

pub fn countDomains(database: *db.Db) db.Error!i64 {
    return database.queryInt("SELECT count(*) FROM domains");
}

// ---------------------------------------------------------------------------
// the API read layer (`GET /api/queries`, `/api/stats`, `/api/stats/timeseries`)
// ---------------------------------------------------------------------------

/// One row of `GET /api/queries`, joined back through the `domains` dimension.
///
/// A summary projection, deliberately narrower than `QueryDetail`: the list is
/// a table the operator scans, and the full provenance of a row is one request
/// away at `GET /api/queries/{id}`.
///
/// The nullable text columns read a NULL as `""` — the same convention
/// `Stmt.columnText` already uses. None of them is ever written as an empty
/// string, so the mapping loses nothing and the API layer can treat `""` as
/// "absent".
pub const QueryRow = struct {
    id: i64,
    ts: i64,
    domain: []const u8,
    client_ip: []const u8,
    qtype: ?u16,
    qclass: u16,
    rcode: u12,
    blocked: bool,
    response_time_us: ?i64,
    cache_hit: ?bool,
    upstream: []const u8,
    policy_action: provenance.PolicyAction,
    policy_reason: provenance.PolicyReason,
    route_kind: provenance.RouteKind,
};

/// Everything one `query_log` row records about one query, for
/// `GET /api/queries/{id}`.
///
/// Same NULL-reads-as-`""` convention as `QueryRow`, and the same closed enums:
/// a stored value the schema does not define is `error.Mismatch`, never passed
/// through as text.
pub const QueryDetail = struct {
    id: i64,
    ts: i64,
    domain: []const u8,
    client_ip: []const u8,
    qtype: ?u16,
    qclass: u16,
    rcode: u12,
    blocked: bool,
    response_time_us: ?i64,
    cache_hit: ?bool,
    upstream: []const u8,
    group_id: ?i64,
    group_name: []const u8,
    policy_action: provenance.PolicyAction,
    policy_reason: provenance.PolicyReason,
    matched: []const u8,
    source_id: ?i64,
    source_name: []const u8,
    cname_target: []const u8,
    safe_search_target: []const u8,
    route_kind: provenance.RouteKind,
    forward_zone: []const u8,
};

/// Every field is an independent narrowing; `null` means "do not filter on it".
///
/// `since` is inclusive and `until` is exclusive, so adjacent windows tile
/// without double-counting a row on the boundary.
pub const QueryFilter = struct {
    limit: u32 = 100,
    /// Keyset cursor: only rows with a strictly smaller `id`. Rows come back
    /// newest-first, so this is the id of the last row of the previous page.
    before: ?i64 = null,
    /// Matched case-insensitively for ASCII, which is what SQLite's `LIKE`
    /// does and what a domain search wants.
    domain_substring: ?[]const u8 = null,
    client: ?[]const u8 = null,
    blocked: ?bool = null,
    since: ?i64 = null,
    until: ?i64 = null,
};

/// Ruling 11 caps the page at 1000; the repository enforces it too, so a caller
/// that forgets cannot ask this connection for the whole table.
pub const max_limit: u32 = 1000;

const select_head =
    \\SELECT q.id, q.timestamp, d.domain, q.client_ip, q.qtype, q.blocked,
    \\       q.response_time_us, q.cache_hit, q.upstream, q.qclass, q.rcode,
    \\       q.policy_action, q.policy_reason, q.route_kind
    \\  FROM query_log q JOIN domains d ON d.id = q.domain_id
;

/// The escape character of `where_domain`. SQLite does not give string literals
/// C escapes, so `'\'` in the SQL text is one backslash.
const like_escape = '\\';

const where_before = " q.id < ?";
const where_domain = " d.domain LIKE ? ESCAPE '\\'";
const where_client = " q.client_ip = ?";
const where_blocked = " q.blocked = ?";
const where_since = " q.timestamp >= ?";
const where_until = " q.timestamp < ?";
const select_tail = " ORDER BY q.id DESC LIMIT ?";

const where_keyword = " WHERE";
const and_keyword = " AND";

/// Assembles the statement from the fixed fragments above and nothing else.
///
/// **No value ever reaches this buffer.** Every filter contributes a `?` and is
/// bound afterwards, in the order the predicates were appended: an unnumbered
/// parameter takes the next free index, so append order and bind order are the
/// same single contract.
const Sql = struct {
    /// `where_keyword` is longer than `and_keyword` and is used at most once,
    /// so counting six of it bounds every reachable combination.
    const capacity = select_head.len + 6 * where_keyword.len + select_tail.len +
        where_before.len + where_domain.len + where_client.len +
        where_blocked.len + where_since.len + where_until.len;

    buf: [capacity]u8 = undefined,
    len: usize = 0,
    has_where: bool = false,

    fn put(self: *Sql, fragment: []const u8) void {
        @memcpy(self.buf[self.len..][0..fragment.len], fragment);
        self.len += fragment.len;
    }

    fn predicate(self: *Sql, fragment: []const u8) void {
        self.put(if (self.has_where) and_keyword else where_keyword);
        self.has_where = true;
        self.put(fragment);
    }

    fn text(self: *const Sql) []const u8 {
        return self.buf[0..self.len];
    }
};

/// Rows come back newest-first (`id DESC`). Every string is allocated from
/// `arena`, including the list's own storage, so the caller frees the whole
/// result by resetting the arena — there is nothing to unwind on failure.
pub fn selectQueries(database: *db.Db, arena: Allocator, filter: QueryFilter) db.Error!std.ArrayList(QueryRow) {
    var sql: Sql = .{};
    sql.put(select_head);
    if (filter.before != null) sql.predicate(where_before);
    if (filter.domain_substring != null) sql.predicate(where_domain);
    if (filter.client != null) sql.predicate(where_client);
    if (filter.blocked != null) sql.predicate(where_blocked);
    if (filter.since != null) sql.predicate(where_since);
    if (filter.until != null) sql.predicate(where_until);
    sql.put(select_tail);

    var stmt = try database.prepare(sql.text());
    defer stmt.deinit();

    var idx: c_int = 0;
    if (filter.before) |v| {
        idx += 1;
        try stmt.bindInt(idx, v);
    }
    if (filter.domain_substring) |v| {
        idx += 1;
        try stmt.bindText(idx, try likePattern(arena, v));
    }
    if (filter.client) |v| {
        idx += 1;
        try stmt.bindText(idx, v);
    }
    if (filter.blocked) |v| {
        idx += 1;
        try stmt.bindBool(idx, v);
    }
    if (filter.since) |v| {
        idx += 1;
        try stmt.bindInt(idx, v);
    }
    if (filter.until) |v| {
        idx += 1;
        try stmt.bindInt(idx, v);
    }
    idx += 1;
    try stmt.bindInt(idx, @min(filter.limit, max_limit));

    var out: std.ArrayList(QueryRow) = .empty;
    while (try stmt.step()) {
        try out.append(arena, .{
            .id = stmt.columnInt(0),
            .ts = stmt.columnInt(1),
            .domain = try stmt.columnTextAlloc(arena, 2),
            .client_ip = try stmt.columnTextAlloc(arena, 3),
            .qtype = if (stmt.isNull(4)) null else std.math.cast(u16, stmt.columnInt(4)) orelse
                return error.Mismatch,
            .blocked = stmt.columnBool(5),
            .response_time_us = if (stmt.isNull(6)) null else stmt.columnInt(6),
            .cache_hit = if (stmt.isNull(7)) null else stmt.columnBool(7),
            .upstream = try stmt.columnTextAlloc(arena, 8),
            .qclass = try columnU16(&stmt, 9),
            .rcode = try columnU12(&stmt, 10),
            .policy_action = try provenance.parse(provenance.PolicyAction, stmt.columnText(11)),
            .policy_reason = try provenance.parse(provenance.PolicyReason, stmt.columnText(12)),
            .route_kind = try provenance.parse(provenance.RouteKind, stmt.columnText(13)),
        });
    }
    return out;
}

/// A `NOT NULL` integer column that the schema bounds to 16 bits. A value
/// outside that range means the row came from something other than this schema.
fn columnU16(stmt: *db.Stmt, col: c_int) db.Error!u16 {
    return std.math.cast(u16, stmt.columnInt(col)) orelse error.Mismatch;
}

/// The `rcode` column, which the schema's `CHECK` bounds to twelve bits. This
/// build cannot write a wider value — the field is a `u12` all the way from the
/// handler — so a row that carries one was written by something else, and is
/// `error.Mismatch` rather than a value truncated into shape.
fn columnU12(stmt: *db.Stmt, col: c_int) db.Error!u12 {
    return std.math.cast(u12, stmt.columnInt(col)) orelse error.Mismatch;
}

const select_detail_sql =
    \\SELECT q.id, q.timestamp, d.domain, q.client_ip, q.qtype, q.blocked,
    \\       q.response_time_us, q.cache_hit, q.upstream, q.qclass, q.rcode,
    \\       q.group_id, q.group_name, q.policy_action, q.policy_reason,
    \\       q.matched, q.source_id, q.source_name, q.cname_target,
    \\       q.safe_search_target, q.route_kind, q.forward_zone
    \\  FROM query_log q JOIN domains d ON d.id = q.domain_id
    \\ WHERE q.id = ?1
;

/// One row's full provenance, or `null` when no row has that id — which is what
/// an id the operator kept from before a retention pass looks like, and is a
/// 404 rather than an error.
///
/// Every string is allocated from `arena`, on the same terms as
/// `selectQueries`.
pub fn detailById(database: *db.Db, arena: Allocator, id: i64) db.Error!?QueryDetail {
    var stmt = try database.prepare(select_detail_sql);
    defer stmt.deinit();
    try stmt.bindInt(1, id);
    if (!try stmt.step()) return null;

    return .{
        .id = stmt.columnInt(0),
        .ts = stmt.columnInt(1),
        .domain = try stmt.columnTextAlloc(arena, 2),
        .client_ip = try stmt.columnTextAlloc(arena, 3),
        .qtype = if (stmt.isNull(4)) null else try columnU16(&stmt, 4),
        .blocked = stmt.columnBool(5),
        .response_time_us = if (stmt.isNull(6)) null else stmt.columnInt(6),
        .cache_hit = if (stmt.isNull(7)) null else stmt.columnBool(7),
        .upstream = try stmt.columnTextAlloc(arena, 8),
        .qclass = try columnU16(&stmt, 9),
        .rcode = try columnU12(&stmt, 10),
        .group_id = if (stmt.isNull(11)) null else stmt.columnInt(11),
        .group_name = try stmt.columnTextAlloc(arena, 12),
        .policy_action = try provenance.parse(provenance.PolicyAction, stmt.columnText(13)),
        .policy_reason = try provenance.parse(provenance.PolicyReason, stmt.columnText(14)),
        .matched = try stmt.columnTextAlloc(arena, 15),
        .source_id = if (stmt.isNull(16)) null else stmt.columnInt(16),
        .source_name = try stmt.columnTextAlloc(arena, 17),
        .cname_target = try stmt.columnTextAlloc(arena, 18),
        .safe_search_target = try stmt.columnTextAlloc(arena, 19),
        .route_kind = try provenance.parse(provenance.RouteKind, stmt.columnText(20)),
        .forward_zone = try stmt.columnTextAlloc(arena, 21),
    };
}

/// Wraps `needle` in `%` and neutralises the two `LIKE` metacharacters, so a
/// user searching for `a_b` gets domains containing `a_b` and not domains
/// containing `axb`. The escape character escapes itself.
fn likePattern(arena: Allocator, needle: []const u8) Allocator.Error![]const u8 {
    var out: std.ArrayList(u8) = try .initCapacity(arena, needle.len * 2 + 2);
    out.appendAssumeCapacity('%');
    for (needle) |ch| {
        if (ch == '%' or ch == '_' or ch == like_escape) out.appendAssumeCapacity(like_escape);
        out.appendAssumeCapacity(ch);
    }
    out.appendAssumeCapacity('%');
    return out.items;
}

/// The `/api/stats` rollup for one period. `avg_response_time_us` is `null` when
/// no row in the window recorded a response time.
pub const StatsTotals = struct {
    queries: u64,
    blocked: u64,
    distinct_clients: u64,
    avg_response_time_us: ?i64,
};

/// The mean is derived from a sum and a count rather than SQL's `avg`, which
/// returns REAL: `Stmt` reads integers, and integer microseconds are exact.
const stats_totals_sql =
    \\SELECT count(*),
    \\       coalesce(sum(blocked <> 0), 0),
    \\       count(DISTINCT client_ip),
    \\       coalesce(sum(response_time_us), 0),
    \\       count(response_time_us)
    \\  FROM query_log
    \\ WHERE timestamp >= ?1 AND timestamp < ?2
;

/// Aggregates `[since, until)`. An empty window is all zeros with a null mean,
/// not an error.
pub fn statsTotals(database: *db.Db, since: i64, until: i64) db.Error!StatsTotals {
    var stmt = try database.prepare(stats_totals_sql);
    defer stmt.deinit();
    try stmt.bindInt(1, since);
    try stmt.bindInt(2, until);

    // A bare aggregate always produces exactly one row; no row means the
    // statement is not the one this function prepared.
    if (!try stmt.step()) return error.Misuse;

    const timed = stmt.columnInt(4);
    return .{
        .queries = try countOf(stmt.columnInt(0)),
        .blocked = try countOf(stmt.columnInt(1)),
        .distinct_clients = try countOf(stmt.columnInt(2)),
        .avg_response_time_us = if (timed == 0) null else @divTrunc(stmt.columnInt(3), timed),
    };
}

/// `count` and `sum` over non-negative columns cannot go negative; a negative
/// value means the row came from something other than this schema.
fn countOf(value: i64) db.Error!u64 {
    if (value < 0) return error.Mismatch;
    return @intCast(value);
}

/// One bucket of `/api/stats/timeseries`. `ts` is the bucket's inclusive start.
pub const Bucket = struct {
    ts: i64,
    queries: u64,
    blocked: u64,
    cached: u64,
};

const timeseries_sql =
    \\SELECT (timestamp - ?1) / ?2,
    \\       count(*),
    \\       coalesce(sum(blocked <> 0), 0),
    \\       coalesce(sum(cache_hit = 1), 0)
    \\  FROM query_log
    \\ WHERE timestamp >= ?1 AND timestamp < ?3
    \\ GROUP BY 1
;

/// Fills `out` with `out.len` buckets of `bucket_seconds` each, covering
/// `[since, since + bucket_seconds * out.len)`, and returns how many it wrote.
///
/// Every bucket is present: a window with no rows in it is written with zeros
/// rather than skipped, so the caller charts a contiguous axis without
/// reconstructing the gaps. Buckets are aligned to `since`, so the caller —
/// which knows the period grammar of ruling 13 — owns UTC alignment by choosing
/// `since`.
pub fn timeseries(database: *db.Db, since: i64, bucket_seconds: u32, out: []Bucket) db.Error!usize {
    if (out.len == 0) return 0;
    // Both are caller bugs, not runtime conditions: a zero width would make the
    // SQL divide by zero (SQLite yields NULL, silently emptying the chart), and
    // a window that does not fit i64 cannot be asked about.
    if (bucket_seconds == 0) return error.Misuse;
    const width: i64 = bucket_seconds;
    const span = std.math.mul(i64, width, std.math.cast(i64, out.len) orelse
        return error.Misuse) catch return error.Misuse;
    const until = std.math.add(i64, since, span) catch return error.Misuse;

    for (out, 0..) |*bucket, i| {
        bucket.* = .{ .ts = since + width * @as(i64, @intCast(i)), .queries = 0, .blocked = 0, .cached = 0 };
    }

    var stmt = try database.prepare(timeseries_sql);
    defer stmt.deinit();
    try stmt.bindInt(1, since);
    try stmt.bindInt(2, width);
    try stmt.bindInt(3, until);

    while (try stmt.step()) {
        // The WHERE clause already bounds the index to `out`; the check is
        // cheap and keeps a schema surprise from writing past the slice.
        const index = std.math.cast(usize, stmt.columnInt(0)) orelse return error.Mismatch;
        if (index >= out.len) return error.Mismatch;
        out[index].queries = try countOf(stmt.columnInt(1));
        out[index].blocked = try countOf(stmt.columnInt(2));
        out[index].cached = try countOf(stmt.columnInt(3));
    }
    return out.len;
}

/// One row of `/api/stats/types`. `qtype` is nullable in the schema, so the
/// rows that carry no type group into a row of their own rather than
/// disappearing from a breakdown that claims to add up.
///
/// No name field: the only qtype-name table lives in the admin, and a second
/// copy here would drift out of agreement with it.
pub const TypeCount = struct {
    qtype: ?u16,
    count: u64,
};

/// `qtype IS NULL` sorts 0 before 1, which puts the null row last within a tie.
/// The ordering is total, so two reads of one window return the same list in
/// the same order — which is what makes the goldens byte-stable. It is an
/// order, not an identity: a caller keys on the `qtype` value, never on a row's
/// position, because a rank change between refreshes moves rows and must not
/// move what they mean.
const stats_types_sql =
    \\SELECT qtype, count(*)
    \\  FROM query_log
    \\ WHERE timestamp >= ?1 AND timestamp < ?2
    \\ GROUP BY qtype
    \\ ORDER BY count(*) DESC, qtype IS NULL, qtype ASC
;

/// The query-type breakdown of `[since, until)`. No zero rows: a type absent
/// from the window is absent from the list.
pub fn statsTypes(
    database: *db.Db,
    arena: Allocator,
    since: i64,
    until: i64,
) db.Error!std.ArrayList(TypeCount) {
    var out: std.ArrayList(TypeCount) = .empty;

    var stmt = try database.prepare(stats_types_sql);
    defer stmt.deinit();
    try stmt.bindInt(1, since);
    try stmt.bindInt(2, until);

    while (try stmt.step()) {
        try out.append(arena, .{
            .qtype = if (stmt.isNull(0)) null else try columnU16(&stmt, 0),
            .count = try countOf(stmt.columnInt(1)),
        });
    }
    return out;
}

/// One row of `/api/stats/routes`: how a slice of the window was answered.
///
/// `source` is the answering resolver's identity and nothing else — the
/// upstream url for `upstream` rows, the zone for `forward_zone` rows, null
/// everywhere else. It is deliberately not `source_name`, which names the
/// blocklist a block came from and would read as an upstream here.
pub const RouteCount = struct {
    route: provenance.RouteKind,
    source: ?[]const u8,
    count: u64,
};

/// A null upstream on an `upstream` row is its own group, not a dropped row: it
/// is a real state of the log and the caller labels it.
const stats_routes_sql =
    \\SELECT route_kind,
    \\       CASE route_kind
    \\            WHEN 'upstream' THEN upstream
    \\            WHEN 'forward_zone' THEN forward_zone
    \\       END AS source,
    \\       count(*)
    \\  FROM query_log
    \\ WHERE timestamp >= ?1 AND timestamp < ?2
    \\ GROUP BY route_kind, source
    \\ ORDER BY count(*) DESC, route_kind ASC, source IS NULL, source ASC
;

/// The answering-route breakdown of `[since, until)`. Strings are copied into
/// `arena`, which outlives the statement.
pub fn statsRoutes(
    database: *db.Db,
    arena: Allocator,
    since: i64,
    until: i64,
) db.Error!std.ArrayList(RouteCount) {
    var out: std.ArrayList(RouteCount) = .empty;

    var stmt = try database.prepare(stats_routes_sql);
    defer stmt.deinit();
    try stmt.bindInt(1, since);
    try stmt.bindInt(2, until);

    while (try stmt.step()) {
        try out.append(arena, .{
            .route = try provenance.parse(provenance.RouteKind, stmt.columnText(0)),
            .source = try stmt.columnTextAllocOrNull(arena, 1),
            .count = try countOf(stmt.columnInt(2)),
        });
    }
    return out;
}

/// How many clients `/api/stats/clients` names before the rest become `other`.
/// Eight is what one legend can carry without becoming a second table.
pub const max_client_series = 8;

/// One named client's series. `buckets` is always the caller's bucket count
/// long, zero-filled, and aligned exactly like `timeseries`.
pub const ClientSeries = struct {
    client: []const u8,
    buckets: []const u64,
};

/// `other` is always present and always bucket-count sized — including for an
/// empty window and for a window with eight clients or fewer. A caller charting
/// a stack must not have to invent the residual series.
pub const ClientsBreakdown = struct {
    clients: []const ClientSeries,
    other: []const u64,
};

/// Ranked by in-window total, ties broken by address, so the cut at eight is
/// the same cut on every request over the same data.
const stats_clients_rank_sql =
    \\SELECT client_ip
    \\  FROM query_log
    \\ WHERE timestamp >= ?1 AND timestamp < ?2
    \\ GROUP BY client_ip
    \\ ORDER BY count(*) DESC, client_ip ASC
    \\ LIMIT ?3
;

const stats_clients_buckets_sql =
    \\SELECT client_ip, (timestamp - ?1) / ?2, count(*)
    \\  FROM query_log
    \\ WHERE timestamp >= ?1 AND timestamp < ?3
    \\ GROUP BY 1, 2
;

/// Per-client counts over `bucket_count` buckets of `bucket_seconds` starting
/// at `since`. Everything outside the top `max_client_series` sums into
/// `other`, so the series still add up to the window's total.
///
/// Two statements, one ranking and one bucketing: the caller runs them inside
/// one read transaction, so the rank and the buckets describe one state.
pub fn statsClients(
    database: *db.Db,
    arena: Allocator,
    since: i64,
    bucket_seconds: u32,
    bucket_count: u32,
) db.Error!ClientsBreakdown {
    if (bucket_seconds == 0 or bucket_count == 0) return error.Misuse;
    const width: i64 = bucket_seconds;
    const span = std.math.mul(i64, width, bucket_count) catch return error.Misuse;
    const until = std.math.add(i64, since, span) catch return error.Misuse;

    var names: std.ArrayList([]const u8) = .empty;
    var series: std.ArrayList([]u64) = .empty;
    {
        var stmt = try database.prepare(stats_clients_rank_sql);
        defer stmt.deinit();
        try stmt.bindInt(1, since);
        try stmt.bindInt(2, until);
        try stmt.bindInt(3, max_client_series);

        while (try stmt.step()) {
            try names.append(arena, try stmt.columnTextAlloc(arena, 0));
            try series.append(arena, try zeroedBuckets(arena, bucket_count));
        }
    }

    const other = try zeroedBuckets(arena, bucket_count);

    var stmt = try database.prepare(stats_clients_buckets_sql);
    defer stmt.deinit();
    try stmt.bindInt(1, since);
    try stmt.bindInt(2, width);
    try stmt.bindInt(3, until);

    while (try stmt.step()) {
        // The WHERE clause bounds the index already; the check keeps a schema
        // surprise from writing past the slice.
        const index = std.math.cast(usize, stmt.columnInt(1)) orelse return error.Mismatch;
        if (index >= bucket_count) return error.Mismatch;
        const count = try countOf(stmt.columnInt(2));

        const client = stmt.columnText(0);
        const target = for (names.items, series.items) |name_, buckets| {
            if (std.mem.eql(u8, name_, client)) break buckets;
        } else other;
        target[index] += count;
    }

    const clients = try arena.alloc(ClientSeries, names.items.len);
    for (clients, names.items, series.items) |*entry, name_, buckets| {
        entry.* = .{ .client = name_, .buckets = buckets };
    }
    return .{ .clients = clients, .other = other };
}

fn zeroedBuckets(arena: Allocator, bucket_count: u32) Allocator.Error![]u64 {
    const buckets = try arena.alloc(u64, bucket_count);
    @memset(buckets, 0);
    return buckets;
}

// ---------------------------------------------------------------------------
// tests
// ---------------------------------------------------------------------------

const logger = @import("../logger.zig");
const querylog_schema = @import("../querylog_schema.zig");

const testing = std.testing;

fn openLog() !db.Db {
    var database = try db.Db.open(":memory:", .{ .mode = .memory });
    errdefer database.close();
    try db.applyPragmas(&database, .{});
    try database.exec(querylog_schema.ddl);
    return database;
}

fn plainRow(timestamp: i64, domain: []const u8) Row {
    return .{
        .timestamp = timestamp,
        .domain = domain,
        .client_ip = "192.0.2.10",
        .qtype = 1,
        .qclass = 1,
        .rcode = 0,
        .blocked = false,
        .response_time_us = 1200,
        .cache_hit = false,
        .upstream = "9.9.9.9",
        .group_id = 1,
        .group_name = "default",
        .policy_action = .allow,
        .policy_reason = .no_match,
        .matched = null,
        .source_id = null,
        .source_name = null,
        .cname_target = null,
        .safe_search_target = null,
        .route_kind = .upstream,
        .forward_zone = null,
    };
}

fn domainIdOf(database: *db.Db, domain: []const u8) !i64 {
    var stmt = try database.prepare(select_domain_sql);
    defer stmt.deinit();
    try stmt.bindText(1, domain);
    try testing.expect(try stmt.step());
    return stmt.columnInt(0);
}

test "a foreign row with an rcode wider than twelve bits is refused, not truncated" {
    var database = try db.Db.open(":memory:", .{ .mode = .memory });
    defer database.close();
    try db.applyPragmas(&database, .{});

    // The shipped schema's `CHECK` makes this row impossible in a file this
    // build created, so the table is built without it. The read path's job is
    // to refuse a `querylog.db` that came from somewhere else rather than to
    // narrow a value it cannot represent.
    try database.exec(
        \\CREATE TABLE domains (id INTEGER PRIMARY KEY, domain TEXT NOT NULL UNIQUE);
        \\CREATE TABLE query_log (
        \\    id INTEGER PRIMARY KEY, timestamp INTEGER NOT NULL,
        \\    domain_id INTEGER NOT NULL, client_ip TEXT NOT NULL,
        \\    qtype INTEGER, blocked INTEGER NOT NULL, response_time_us INTEGER,
        \\    cache_hit INTEGER, upstream TEXT, qclass INTEGER NOT NULL,
        \\    rcode INTEGER NOT NULL, group_id INTEGER, group_name TEXT,
        \\    policy_action TEXT NOT NULL, policy_reason TEXT NOT NULL,
        \\    matched TEXT, source_id INTEGER, source_name TEXT,
        \\    cname_target TEXT, safe_search_target TEXT,
        \\    route_kind TEXT NOT NULL, forward_zone TEXT
        \\);
        \\INSERT INTO domains (id, domain) VALUES (1, 'a.example');
        \\INSERT INTO query_log
        \\  (id, timestamp, domain_id, client_ip, blocked, qclass, rcode,
        \\   policy_action, policy_reason, route_kind)
        \\VALUES (1, 10, 1, '192.0.2.10', 0, 1, 4096, 'allow', 'no_match', 'upstream');
    );

    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();
    const arena = arena_state.allocator();

    try testing.expectError(error.Mismatch, selectQueries(&database, arena, .{}));
    try testing.expectError(error.Mismatch, detailById(&database, arena, 1));
}

test "writeBatch inserts every row and interns each domain once" {
    var database = try openLog();
    defer database.close();
    var writer = try BatchWriter.init(&database);
    defer writer.deinit();

    try writer.writeBatch(&.{
        plainRow(100, "example.com"),
        plainRow(101, "example.com"),
        plainRow(102, "ads.example.net"),
    });

    try testing.expectEqual(@as(i64, 3), try countRows(&database));
    try testing.expectEqual(@as(i64, 2), try countDomains(&database));

    const first = try domainIdOf(&database, "example.com");
    try testing.expectEqual(
        @as(i64, 2),
        try database.queryInt("SELECT count(*) FROM query_log WHERE domain_id = 1"),
    );
    try testing.expectEqual(@as(i64, 1), first);
}

test "a second batch reuses the interned domain id" {
    var database = try openLog();
    defer database.close();
    var writer = try BatchWriter.init(&database);
    defer writer.deinit();

    try writer.writeBatch(&.{plainRow(100, "example.com")});
    const before = try domainIdOf(&database, "example.com");

    try writer.writeBatch(&.{ plainRow(200, "example.com"), plainRow(201, "other.example") });
    const after = try domainIdOf(&database, "example.com");

    try testing.expectEqual(before, after);
    try testing.expectEqual(@as(i64, 3), try countRows(&database));
    try testing.expectEqual(@as(i64, 2), try countDomains(&database));
    try testing.expectEqual(
        @as(i64, 2),
        try database.queryInt("SELECT count(*) FROM query_log WHERE domain_id = 1"),
    );
}

test "every column round-trips a value and a null" {
    var database = try openLog();
    defer database.close();
    var writer = try BatchWriter.init(&database);
    defer writer.deinit();

    try writer.writeBatch(&.{
        .{
            .timestamp = 10,
            .domain = "blocked.example",
            .client_ip = "2001:db8::1",
            .qtype = 28,
            .qclass = 1,
            .rcode = 3,
            .blocked = true,
            .response_time_us = 42,
            .cache_hit = true,
            .upstream = "https://dns.example/dns-query",
            .group_id = 7,
            .group_name = "kids",
            .policy_action = .block,
            .policy_reason = .blocklist_wildcard,
            .matched = "*.ads.example",
            .source_id = 3,
            .source_name = "steven black",
            .cname_target = "tracker.cdn.example",
            .safe_search_target = "forcesafesearch.google.com",
            .route_kind = .blocked,
            .forward_zone = "home.arpa",
        },
        // Every nullable column absent at once, which is the shape of a query
        // the pipeline answered before any of them applied.
        .{
            .timestamp = 11,
            .domain = "quiet.example",
            .client_ip = "hidden",
            .qtype = null,
            .qclass = 3,
            .rcode = 0,
            .blocked = false,
            .response_time_us = null,
            .cache_hit = null,
            .upstream = null,
            .group_id = null,
            .group_name = null,
            .policy_action = .not_evaluated,
            .policy_reason = .non_in_class,
            .matched = null,
            .source_id = null,
            .source_name = null,
            .cname_target = null,
            .safe_search_target = null,
            .route_kind = .upstream,
            .forward_zone = null,
        },
    });

    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();
    const arena = arena_state.allocator();

    const full = (try detailById(&database, arena, 1)).?;
    try testing.expectEqualStrings("blocked.example", full.domain);
    try testing.expectEqualStrings("2001:db8::1", full.client_ip);
    try testing.expectEqual(@as(?u16, 28), full.qtype);
    try testing.expectEqual(@as(u16, 1), full.qclass);
    try testing.expectEqual(@as(u12, 3), full.rcode);
    try testing.expect(full.blocked);
    try testing.expectEqual(@as(?i64, 42), full.response_time_us);
    try testing.expectEqual(@as(?bool, true), full.cache_hit);
    try testing.expectEqualStrings("https://dns.example/dns-query", full.upstream);
    try testing.expectEqual(@as(?i64, 7), full.group_id);
    try testing.expectEqualStrings("kids", full.group_name);
    try testing.expectEqual(provenance.PolicyAction.block, full.policy_action);
    try testing.expectEqual(provenance.PolicyReason.blocklist_wildcard, full.policy_reason);
    try testing.expectEqualStrings("*.ads.example", full.matched);
    try testing.expectEqual(@as(?i64, 3), full.source_id);
    try testing.expectEqualStrings("steven black", full.source_name);
    try testing.expectEqualStrings("tracker.cdn.example", full.cname_target);
    try testing.expectEqualStrings("forcesafesearch.google.com", full.safe_search_target);
    try testing.expectEqual(provenance.RouteKind.blocked, full.route_kind);
    try testing.expectEqualStrings("home.arpa", full.forward_zone);

    // A NULL text column reads as the empty string, by the documented
    // convention; a NULL integer stays null, because 0 is a real id.
    const bare = (try detailById(&database, arena, 2)).?;
    try testing.expectEqual(@as(?u16, null), bare.qtype);
    try testing.expectEqual(@as(u16, 3), bare.qclass);
    try testing.expectEqual(@as(?i64, null), bare.response_time_us);
    try testing.expectEqual(@as(?bool, null), bare.cache_hit);
    try testing.expectEqualStrings("", bare.upstream);
    try testing.expectEqual(@as(?i64, null), bare.group_id);
    try testing.expectEqualStrings("", bare.group_name);
    try testing.expectEqual(provenance.PolicyAction.not_evaluated, bare.policy_action);
    try testing.expectEqual(provenance.PolicyReason.non_in_class, bare.policy_reason);
    try testing.expectEqualStrings("", bare.matched);
    try testing.expectEqual(@as(?i64, null), bare.source_id);
    try testing.expectEqualStrings("", bare.source_name);
    try testing.expectEqualStrings("", bare.cname_target);
    try testing.expectEqualStrings("", bare.safe_search_target);
    try testing.expectEqual(provenance.RouteKind.upstream, bare.route_kind);
    try testing.expectEqualStrings("", bare.forward_zone);
}

test "detailById returns null for an id no row has" {
    var database = try openLog();
    defer database.close();
    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();

    try seed(&database, &.{plainRow(10, "a.example")});

    // An id from before a retention pass looks exactly like this, and is a 404
    // rather than an error.
    try testing.expectEqual(@as(?QueryDetail, null), try detailById(&database, arena_state.allocator(), 2));
    try testing.expectEqual(@as(?QueryDetail, null), try detailById(&database, arena_state.allocator(), 0));
    try testing.expect((try detailById(&database, arena_state.allocator(), 1)) != null);
}

test "a stored enum value the schema does not define is a data error, not a passthrough" {
    var database = try openLog();
    defer database.close();
    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();
    const arena = arena_state.allocator();

    try seed(&database, &.{plainRow(10, "a.example")});
    try database.exec("UPDATE query_log SET policy_reason = 'whatever' WHERE id = 1;");

    try testing.expectError(error.Mismatch, detailById(&database, arena, 1));
    try testing.expectError(error.Mismatch, selectQueries(&database, arena, .{}));
}

test "an empty batch writes nothing and opens no transaction" {
    var database = try openLog();
    defer database.close();
    var writer = try BatchWriter.init(&database);
    defer writer.deinit();

    // A transaction is already open, so a `BEGIN IMMEDIATE` from `writeBatch`
    // would fail: this is what proves the empty batch returns before it.
    var tx = try db.Tx.begin(&database);
    try writer.writeBatch(&.{});
    tx.rollback();

    try testing.expectEqual(@as(i64, 0), try countRows(&database));
    try testing.expectEqual(@as(i64, 0), try countDomains(&database));
}

test "pruneOlderThan deletes strictly older rows and returns the count" {
    var database = try openLog();
    defer database.close();
    var writer = try BatchWriter.init(&database);
    defer writer.deinit();

    try writer.writeBatch(&.{
        plainRow(100, "old.example"),
        plainRow(199, "old.example"),
        plainRow(200, "edge.example"),
        plainRow(300, "fresh.example"),
    });

    try testing.expectEqual(@as(i64, 2), (try pruneOlderThan(&database, 200)).deleted);
    try testing.expectEqual(@as(i64, 2), try countRows(&database));
    // The row exactly at the cutoff stays.
    try testing.expectEqual(
        @as(i64, 1),
        try database.queryInt("SELECT count(*) FROM query_log WHERE timestamp = 200"),
    );
    // A second pass over the same cutoff finds nothing left to do.
    try testing.expectEqual(@as(i64, 0), (try pruneOlderThan(&database, 200)).deleted);
}

test "a prune advances the coverage watermark to its own cutoff" {
    var database = try openLog();
    defer database.close();
    var writer = try BatchWriter.init(&database);
    defer writer.deinit();

    // The seeded watermark is `created_at + 1`, which is now-ish; the cutoffs
    // below are all in the past, so they start out behind it.
    const start = try availableSince(&database);
    try writer.writeBatch(&.{ plainRow(start + 100, "a.example"), plainRow(start + 300, "b.example") });

    const first = try pruneOlderThan(&database, start + 200);
    try testing.expectEqual(@as(i64, 1), first.deleted);
    try testing.expectEqual(start + 200, first.available_since);
    try testing.expectEqual(start + 200, try availableSince(&database));

    // A prune that deletes nothing still advances: the window it swept is
    // covered whether or not it held rows.
    const second = try pruneOlderThan(&database, start + 250);
    try testing.expectEqual(@as(i64, 0), second.deleted);
    try testing.expectEqual(start + 250, try availableSince(&database));
}

test "the watermark never moves backward" {
    var database = try openLog();
    defer database.close();

    const start = try availableSince(&database);
    const advanced = try pruneOlderThan(&database, start + 1000);
    try testing.expectEqual(start + 1000, advanced.available_since);

    // A shortened `retention_days`, a clock that stepped back, a pass with a
    // stale cutoff: none of them may widen the promise the file makes.
    for ([_]i64{ start + 999, start, start - 100_000, 0 }) |older| {
        const result = try pruneOlderThan(&database, older);
        try testing.expectEqual(start + 1000, result.available_since);
        try testing.expectEqual(start + 1000, try availableSince(&database));
    }
}

test "a failed delete leaves both the rows and the watermark untouched" {
    var database = try openLog();
    defer database.close();
    var writer = try BatchWriter.init(&database);
    defer writer.deinit();

    const start = try availableSince(&database);
    try writer.writeBatch(&.{plainRow(start - 100, "old.example")});
    try database.exec(
        \\CREATE TRIGGER refuse_delete BEFORE DELETE ON query_log
        \\BEGIN SELECT RAISE(ABORT, 'refused'); END;
    );

    try testing.expectError(error.Constraint, pruneOlderThan(&database, start + 1000));

    try testing.expectEqual(@as(i64, 1), try countRows(&database));
    try testing.expectEqual(start, try availableSince(&database));
}

test "a failed watermark update leaves the rows it had already deleted" {
    var database = try openLog();
    defer database.close();
    var writer = try BatchWriter.init(&database);
    defer writer.deinit();

    const start = try availableSince(&database);
    try writer.writeBatch(&.{plainRow(start - 100, "old.example")});
    // The delete succeeds and the advance does not. Without one transaction
    // around the pair, this is the case that loses rows the watermark still
    // promises.
    try database.exec(
        \\CREATE TRIGGER refuse_advance BEFORE UPDATE ON querylog_meta
        \\BEGIN SELECT RAISE(ABORT, 'refused'); END;
    );

    try testing.expectError(error.Constraint, pruneOlderThan(&database, start + 1000));

    try testing.expectEqual(@as(i64, 1), try countRows(&database));
    try testing.expectEqual(start, try availableSince(&database));
}

test "a failed commit rolls back the delete and the watermark together" {
    var database = try openLog();
    defer database.close();
    var writer = try BatchWriter.init(&database);
    defer writer.deinit();

    const start = try availableSince(&database);
    try writer.writeBatch(&.{plainRow(start - 100, "old.example")});

    // Both statements succeed and COMMIT is what fails: the advance inserts a
    // `query_log` row whose `domain_id` references nothing, and
    // `defer_foreign_keys` holds that violation back until the commit checks
    // it (SQLite's documented semantics for the pragma; the assertions below
    // observe the rollback, not the moment the check ran).
    try database.exec(
        \\CREATE TRIGGER break_at_commit AFTER UPDATE ON querylog_meta
        \\BEGIN INSERT INTO query_log
        \\  (timestamp, domain_id, client_ip, blocked, qclass, rcode,
        \\   policy_action, policy_reason, route_kind)
        \\VALUES (1, 999999, 'x', 0, 1, 0, 'allow', 'no_match', 'upstream'); END;
    );
    try database.exec("PRAGMA defer_foreign_keys = ON;");

    try testing.expectError(error.Constraint, pruneOlderThan(&database, start + 1000));

    // Nothing survived: not the delete, not the advance, not the row the
    // trigger inserted.
    try testing.expectEqual(@as(i64, 1), try countRows(&database));
    try testing.expectEqual(start, try availableSince(&database));
    try testing.expectEqual(
        @as(i64, 0),
        try database.queryInt("SELECT count(*) FROM query_log WHERE client_ip = 'x'"),
    );
}

test "pruneOlderThan leaves the domains dimension table intact" {
    var database = try openLog();
    defer database.close();
    var writer = try BatchWriter.init(&database);
    defer writer.deinit();

    try writer.writeBatch(&.{ plainRow(10, "a.example"), plainRow(11, "b.example") });
    try testing.expectEqual(@as(i64, 2), (try pruneOlderThan(&database, 1000)).deleted);

    try testing.expectEqual(@as(i64, 0), try countRows(&database));
    try testing.expectEqual(@as(i64, 2), try countDomains(&database));
}

test "a failing row rolls the whole batch back and the writer survives it" {
    var database = try openLog();
    defer database.close();
    try database.exec(
        \\CREATE TRIGGER refuse_boom BEFORE INSERT ON query_log
        \\WHEN new.client_ip = 'boom'
        \\BEGIN SELECT RAISE(ABORT, 'refused'); END;
    );

    var writer = try BatchWriter.init(&database);
    defer writer.deinit();

    var doomed = plainRow(20, "second.example");
    doomed.client_ip = "boom";
    try testing.expectError(error.Constraint, writer.writeBatch(&.{
        plainRow(10, "first.example"),
        doomed,
    }));

    // The interned domain of the row that did insert is gone with it.
    try testing.expectEqual(@as(i64, 0), try countRows(&database));
    try testing.expectEqual(@as(i64, 0), try countDomains(&database));

    try writer.writeBatch(&.{plainRow(30, "third.example")});
    try testing.expectEqual(@as(i64, 1), try countRows(&database));
    try testing.expectEqual(@as(i64, 1), try countDomains(&database));
}

test "countRows and countDomains agree with what the batches wrote" {
    var database = try openLog();
    defer database.close();
    var writer = try BatchWriter.init(&database);
    defer writer.deinit();

    try testing.expectEqual(@as(i64, 0), try countRows(&database));
    try testing.expectEqual(@as(i64, 0), try countDomains(&database));

    var rows: [50]Row = undefined;
    var names: [50][16]u8 = undefined;
    for (&rows, &names, 0..) |*row, *name, i| {
        const written = std.fmt.bufPrint(name, "d{d}.example", .{i % 7}) catch unreachable;
        row.* = plainRow(@intCast(i), written);
    }
    try writer.writeBatch(&rows);

    try testing.expectEqual(@as(i64, 50), try countRows(&database));
    try testing.expectEqual(@as(i64, 7), try countDomains(&database));
}

// `PRAGMA wal_checkpoint` needs a real WAL, which an in-memory database cannot
// have. `std.testing.tmpDir` creates its directory under `.zig-cache/tmp/`
// relative to the process working directory, which is also how SQLite's VFS
// resolves the filename it is handed (`storage_integration_test.zig:44`).
const tmp_prefix = ".zig-cache/tmp/";
const sub_path_len = @typeInfo(@FieldType(testing.TmpDir, "sub_path")).array.len;

test "checkpointTruncate and vacuum run against a WAL file database" {
    var tmp = testing.tmpDir(.{});
    defer tmp.cleanup();

    var path_buf: [tmp_prefix.len + sub_path_len + 32]u8 = undefined;
    const path = try std.fmt.bufPrintZ(&path_buf, "{s}{s}/querylog.db", .{ tmp_prefix, &tmp.sub_path });

    var database = try db.Db.open(path, .{ .mode = .read_write_create });
    defer database.close();
    try db.applyPragmas(&database, .{});
    {
        var stmt = try database.prepare("PRAGMA journal_mode");
        defer stmt.deinit();
        try testing.expect(try stmt.step());
        // `columnText` is borrowed until the next call on the statement, so it
        // is compared here rather than carried out of this block.
        try testing.expectEqualStrings("wal", stmt.columnText(0));
    }
    try database.exec(querylog_schema.ddl);

    var writer = try BatchWriter.init(&database);
    defer writer.deinit();
    try writer.writeBatch(&.{ plainRow(10, "a.example"), plainRow(20, "b.example") });

    try checkpointTruncate(&database);
    try testing.expectEqual(@as(i64, 1), (try pruneOlderThan(&database, 20)).deleted);
    try checkpointTruncate(&database);
    try vacuum(&database);

    try testing.expectEqual(@as(i64, 1), try countRows(&database));
    try testing.expectEqual(@as(i64, 2), try countDomains(&database));
}

// --- the read layer -------------------------------------------------------

/// `BatchWriter` assigns `query_log.id` in the order it is handed the rows, so
/// every test below knows the id of each seeded row: the nth row of the nth
/// batch has id n.
fn seed(database: *db.Db, rows: []const Row) !void {
    var writer = try BatchWriter.init(database);
    defer writer.deinit();
    try writer.writeBatch(rows);
}

fn ids(rows: []const QueryRow, out: []i64) []const i64 {
    for (rows, 0..) |row, i| out[i] = row.id;
    return out[0..rows.len];
}

test "selectQueries returns the newest row first and reads every column" {
    var database = try openLog();
    defer database.close();
    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();

    try seed(&database, &.{
        .{
            .timestamp = 10,
            .domain = "ads.example.net",
            .client_ip = "192.0.2.10",
            .qtype = 28,
            .qclass = 1,
            .rcode = 0,
            .blocked = true,
            .response_time_us = 4200,
            .cache_hit = true,
            .upstream = "https://dns.example/dns-query",
            .group_id = 2,
            .group_name = "kids",
            .policy_action = .block,
            .policy_reason = .blocklist_domain,
            .matched = "ads.example.net",
            .source_id = 5,
            .source_name = "steven black",
            .cname_target = null,
            .safe_search_target = null,
            .route_kind = .blocked,
            .forward_zone = null,
        },
        .{
            .timestamp = 20,
            .domain = "quiet.example",
            .client_ip = "hidden",
            .qtype = null,
            .qclass = 1,
            .rcode = 2,
            .blocked = false,
            .response_time_us = null,
            .cache_hit = null,
            .upstream = null,
            .group_id = null,
            .group_name = null,
            .policy_action = .not_evaluated,
            .policy_reason = .snapshot_unavailable,
            .matched = null,
            .source_id = null,
            .source_name = null,
            .cname_target = null,
            .safe_search_target = null,
            .route_kind = .rejected,
            .forward_zone = null,
        },
    });

    const rows = try selectQueries(&database, arena_state.allocator(), .{});
    try testing.expectEqual(@as(usize, 2), rows.items.len);

    const newest = rows.items[0];
    try testing.expectEqual(@as(i64, 2), newest.id);
    try testing.expectEqual(@as(i64, 20), newest.ts);
    try testing.expectEqualStrings("quiet.example", newest.domain);
    try testing.expectEqualStrings("hidden", newest.client_ip);
    try testing.expectEqual(@as(?u16, null), newest.qtype);
    try testing.expectEqual(@as(u16, 1), newest.qclass);
    try testing.expectEqual(@as(u12, 2), newest.rcode);
    try testing.expect(!newest.blocked);
    try testing.expectEqual(@as(?i64, null), newest.response_time_us);
    try testing.expectEqual(@as(?bool, null), newest.cache_hit);
    // A NULL text column reads as the empty string, by documented convention.
    try testing.expectEqualStrings("", newest.upstream);
    try testing.expectEqual(provenance.PolicyAction.not_evaluated, newest.policy_action);
    try testing.expectEqual(provenance.PolicyReason.snapshot_unavailable, newest.policy_reason);
    try testing.expectEqual(provenance.RouteKind.rejected, newest.route_kind);

    const oldest = rows.items[1];
    try testing.expectEqual(@as(i64, 1), oldest.id);
    try testing.expectEqual(@as(i64, 10), oldest.ts);
    try testing.expectEqualStrings("ads.example.net", oldest.domain);
    try testing.expectEqualStrings("192.0.2.10", oldest.client_ip);
    try testing.expectEqual(@as(?u16, 28), oldest.qtype);
    try testing.expectEqual(@as(u16, 1), oldest.qclass);
    try testing.expectEqual(@as(u12, 0), oldest.rcode);
    try testing.expect(oldest.blocked);
    try testing.expectEqual(@as(?i64, 4200), oldest.response_time_us);
    try testing.expectEqual(@as(?bool, true), oldest.cache_hit);
    try testing.expectEqualStrings("https://dns.example/dns-query", oldest.upstream);
    try testing.expectEqual(provenance.PolicyAction.block, oldest.policy_action);
    try testing.expectEqual(provenance.PolicyReason.blocklist_domain, oldest.policy_reason);
    try testing.expectEqual(provenance.RouteKind.blocked, oldest.route_kind);
}

test "selectQueries honours the limit and caps it at max_limit" {
    var database = try openLog();
    defer database.close();
    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();
    const arena = arena_state.allocator();

    var rows: [1005]Row = undefined;
    for (&rows, 0..) |*row, i| row.* = plainRow(@intCast(i), "example.com");
    try seed(&database, &rows);

    const few = try selectQueries(&database, arena, .{ .limit = 3 });
    try testing.expectEqual(@as(usize, 3), few.items.len);

    // Asked for more than the cap, and for more rows than the cap, so the cap
    // is what bounds the answer rather than the table.
    const capped = try selectQueries(&database, arena, .{ .limit = 5000 });
    try testing.expectEqual(@as(usize, max_limit), capped.items.len);

    const none = try selectQueries(&database, arena, .{ .limit = 0 });
    try testing.expectEqual(@as(usize, 0), none.items.len);
}

test "keyset paging walks every row exactly once across the page boundaries" {
    var database = try openLog();
    defer database.close();
    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();
    const arena = arena_state.allocator();

    var seeded: [7]Row = undefined;
    for (&seeded, 0..) |*row, i| row.* = plainRow(@intCast(i), "example.com");
    try seed(&database, &seeded);

    var seen: std.ArrayList(i64) = .empty;
    defer seen.deinit(testing.allocator);

    var before: ?i64 = null;
    var pages: usize = 0;
    while (pages < 10) : (pages += 1) {
        const page = try selectQueries(&database, arena, .{ .limit = 3, .before = before });
        if (page.items.len == 0) break;
        for (page.items) |row| try seen.append(testing.allocator, row.id);
        before = page.items[page.items.len - 1].id;
    }

    // Two full pages and one short page; the fourth call returns nothing and
    // breaks before the counter, which is how the walk knows it is done.
    try testing.expectEqual(@as(usize, 3), pages);
    try testing.expectEqualSlices(i64, &.{ 7, 6, 5, 4, 3, 2, 1 }, seen.items);
}

test "each filter narrows the result on its own" {
    var database = try openLog();
    defer database.close();
    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();
    const arena = arena_state.allocator();

    var blocked_row = plainRow(200, "ads.example.net");
    blocked_row.client_ip = "192.0.2.20";
    blocked_row.blocked = true;
    blocked_row.policy_action = .block;
    blocked_row.policy_reason = .blocklist_domain;
    blocked_row.route_kind = .blocked;
    try seed(&database, &.{
        plainRow(100, "one.example.com"),
        blocked_row,
        plainRow(300, "two.example.com"),
    });

    var buf: [8]i64 = undefined;

    const by_domain = try selectQueries(&database, arena, .{ .domain_substring = "example.com" });
    try testing.expectEqualSlices(i64, &.{ 3, 1 }, ids(by_domain.items, &buf));

    const by_client = try selectQueries(&database, arena, .{ .client = "192.0.2.20" });
    try testing.expectEqualSlices(i64, &.{2}, ids(by_client.items, &buf));

    // An exact match, not a prefix: the seeded clients share the first octets.
    const no_client = try selectQueries(&database, arena, .{ .client = "192.0.2" });
    try testing.expectEqual(@as(usize, 0), no_client.items.len);

    const only_blocked = try selectQueries(&database, arena, .{ .blocked = true });
    try testing.expectEqualSlices(i64, &.{2}, ids(only_blocked.items, &buf));

    const only_allowed = try selectQueries(&database, arena, .{ .blocked = false });
    try testing.expectEqualSlices(i64, &.{ 3, 1 }, ids(only_allowed.items, &buf));

    // Every filter at once, all satisfied by the one blocked row.
    const combined = try selectQueries(&database, arena, .{
        .limit = 10,
        .before = 3,
        .domain_substring = "ads",
        .client = "192.0.2.20",
        .blocked = true,
        .since = 200,
        .until = 300,
    });
    try testing.expectEqualSlices(i64, &.{2}, ids(combined.items, &buf));
}

test "since is inclusive, until is exclusive, and an empty range selects nothing" {
    var database = try openLog();
    defer database.close();
    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();
    const arena = arena_state.allocator();

    try seed(&database, &.{
        plainRow(100, "a.example"),
        plainRow(200, "b.example"),
        plainRow(300, "c.example"),
    });

    var buf: [8]i64 = undefined;

    const window = try selectQueries(&database, arena, .{ .since = 100, .until = 300 });
    try testing.expectEqualSlices(i64, &.{ 2, 1 }, ids(window.items, &buf));

    const after = try selectQueries(&database, arena, .{ .since = 300 });
    try testing.expectEqualSlices(i64, &.{3}, ids(after.items, &buf));

    const empty = try selectQueries(&database, arena, .{ .since = 300, .until = 300 });
    try testing.expectEqual(@as(usize, 0), empty.items.len);

    const beyond = try selectQueries(&database, arena, .{ .since = 1000 });
    try testing.expectEqual(@as(usize, 0), beyond.items.len);
}

test "a domain substring matches % and _ as literal characters" {
    var database = try openLog();
    defer database.close();
    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();
    const arena = arena_state.allocator();

    try seed(&database, &.{
        plainRow(10, "a_b.example"),
        plainRow(20, "axb.example"),
        plainRow(30, "a%b.example"),
        plainRow(40, "azzb.example"),
        plainRow(50, "back\\slash.example"),
    });

    var buf: [8]i64 = undefined;

    // Unescaped, `_` is LIKE's single-character wildcard and would also match
    // "axb"; escaped, it matches only the underscore.
    const underscore = try selectQueries(&database, arena, .{ .domain_substring = "a_b" });
    try testing.expectEqualSlices(i64, &.{1}, ids(underscore.items, &buf));

    // Unescaped, `%` would match everything from "a" to "b", so "azzb" too.
    const percent = try selectQueries(&database, arena, .{ .domain_substring = "a%b" });
    try testing.expectEqualSlices(i64, &.{3}, ids(percent.items, &buf));

    // The escape character escapes itself, so it is searchable as well.
    const backslash = try selectQueries(&database, arena, .{ .domain_substring = "k\\s" });
    try testing.expectEqualSlices(i64, &.{5}, ids(backslash.items, &buf));

    // An empty needle is `%%`, which matches every row rather than none.
    const all = try selectQueries(&database, arena, .{ .domain_substring = "" });
    try testing.expectEqual(@as(usize, 5), all.items.len);
}

test "statsTotals aggregates the window and averages only the timed rows" {
    var database = try openLog();
    defer database.close();

    var timed = plainRow(100, "a.example");
    timed.response_time_us = 100;
    var blocked_row = plainRow(150, "ads.example");
    blocked_row.blocked = true;
    blocked_row.policy_action = .block;
    blocked_row.policy_reason = .blocklist_domain;
    blocked_row.route_kind = .blocked;
    blocked_row.response_time_us = 200;
    var cached = plainRow(199, "b.example");
    cached.client_ip = "192.0.2.99";
    cached.cache_hit = true;
    cached.response_time_us = null;
    try seed(&database, &.{ timed, blocked_row, cached, plainRow(200, "outside.example") });

    const totals = try statsTotals(&database, 100, 200);
    try testing.expectEqual(@as(u64, 3), totals.queries);
    try testing.expectEqual(@as(u64, 1), totals.blocked);
    try testing.expectEqual(@as(u64, 2), totals.distinct_clients);
    // (100 + 200) / 2 — the untimed row is not in the divisor.
    try testing.expectEqual(@as(?i64, 150), totals.avg_response_time_us);
}

test "statsTotals over an empty window is zeros with a null average" {
    var database = try openLog();
    defer database.close();
    try seed(&database, &.{plainRow(100, "a.example")});

    for ([_][2]i64{ .{ 500, 600 }, .{ 100, 100 } }) |window| {
        const totals = try statsTotals(&database, window[0], window[1]);
        try testing.expectEqual(@as(u64, 0), totals.queries);
        try testing.expectEqual(@as(u64, 0), totals.blocked);
        try testing.expectEqual(@as(u64, 0), totals.distinct_clients);
        try testing.expectEqual(@as(?i64, null), totals.avg_response_time_us);
    }
}

test "timeseries writes every bucket, including the ones with no rows" {
    var database = try openLog();
    defer database.close();

    var blocked_row = plainRow(1020, "ads.example");
    blocked_row.blocked = true;
    blocked_row.policy_action = .block;
    blocked_row.policy_reason = .blocklist_domain;
    blocked_row.route_kind = .blocked;
    var cached = plainRow(1035, "b.example");
    cached.cache_hit = true;
    try seed(&database, &.{
        plainRow(995, "before.example"),
        plainRow(1000, "a.example"),
        plainRow(1009, "a.example"),
        blocked_row,
        cached,
        plainRow(1040, "after.example"),
    });

    var buckets: [4]Bucket = undefined;
    try testing.expectEqual(@as(usize, 4), try timeseries(&database, 1000, 10, &buckets));

    // The row at 995 is before the window and the row at 1040 is past its end;
    // neither lands in a bucket.
    try testing.expectEqualSlices(Bucket, &.{
        .{ .ts = 1000, .queries = 2, .blocked = 0, .cached = 0 },
        .{ .ts = 1010, .queries = 0, .blocked = 0, .cached = 0 },
        .{ .ts = 1020, .queries = 1, .blocked = 1, .cached = 0 },
        .{ .ts = 1030, .queries = 1, .blocked = 0, .cached = 1 },
    }, &buckets);
}

test "timeseries over an empty table still writes the whole axis" {
    var database = try openLog();
    defer database.close();

    var buckets: [3]Bucket = undefined;
    try testing.expectEqual(@as(usize, 3), try timeseries(&database, 0, 60, &buckets));
    try testing.expectEqualSlices(Bucket, &.{
        .{ .ts = 0, .queries = 0, .blocked = 0, .cached = 0 },
        .{ .ts = 60, .queries = 0, .blocked = 0, .cached = 0 },
        .{ .ts = 120, .queries = 0, .blocked = 0, .cached = 0 },
    }, &buckets);
}

test "timeseries rejects a zero-width bucket and accepts an empty slice" {
    var database = try openLog();
    defer database.close();

    var buckets: [2]Bucket = undefined;
    try testing.expectError(error.Misuse, timeseries(&database, 0, 0, &buckets));

    var none: [0]Bucket = undefined;
    try testing.expectEqual(@as(usize, 0), try timeseries(&database, 0, 0, &none));
}

test "timeseries reports a window that does not fit an i64 rather than wrapping" {
    var database = try openLog();
    defer database.close();

    var buckets: [4]Bucket = undefined;
    try testing.expectError(
        error.Misuse,
        timeseries(&database, std.math.maxInt(i64) - 1, 3600, &buckets),
    );
}

test "the built SQL never carries a filter value and fits its buffer" {
    var sql: Sql = .{};
    sql.put(select_head);
    sql.predicate(where_before);
    sql.predicate(where_domain);
    sql.predicate(where_client);
    sql.predicate(where_blocked);
    sql.predicate(where_since);
    sql.predicate(where_until);
    sql.put(select_tail);

    // Every predicate present is the longest reachable statement.
    try testing.expect(sql.len <= Sql.capacity);
    try testing.expectEqual(@as(usize, 1), std.mem.count(u8, sql.text(), " WHERE"));
    try testing.expectEqual(@as(usize, 5), std.mem.count(u8, sql.text(), " AND"));
    // Six filters plus the LIMIT, each a bare parameter.
    try testing.expectEqual(@as(usize, 7), std.mem.count(u8, sql.text(), "?"));
}

test "likePattern wraps the needle and neutralises every metacharacter" {
    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();
    const arena = arena_state.allocator();

    try testing.expectEqualStrings("%plain%", try likePattern(arena, "plain"));
    try testing.expectEqualStrings("%a\\_b%", try likePattern(arena, "a_b"));
    try testing.expectEqualStrings("%a\\%b%", try likePattern(arena, "a%b"));
    try testing.expectEqualStrings("%a\\\\b%", try likePattern(arena, "a\\b"));
    try testing.expectEqualStrings("%%", try likePattern(arena, ""));
}

// ---------------------------------------------------------------------------
// period aggregations (milestone 30)
// ---------------------------------------------------------------------------

const agg_since: i64 = 1_700_000_000;
const agg_width: u32 = 60;
const agg_buckets: u32 = 10;
const agg_until: i64 = agg_since + agg_width * agg_buckets;

/// One row of the aggregation fixtures. Everything the three breakdowns read
/// is a parameter; everything else is the same on every row, so a test that
/// changes an outcome names the reason it changed.
fn aggRow(offset: i64, client: []const u8, qtype: ?u16, kind: provenance.RouteKind, source: ?[]const u8) Row {
    return .{
        .timestamp = agg_since + offset,
        .domain = "example.com",
        .client_ip = client,
        .qtype = qtype,
        .qclass = 1,
        .rcode = 0,
        .blocked = kind == .blocked,
        .response_time_us = 1000,
        .cache_hit = kind == .cache,
        .upstream = if (kind == .upstream) source else null,
        .group_id = 1,
        .group_name = "default",
        .policy_action = if (kind == .blocked) .block else .allow,
        .policy_reason = if (kind == .blocked) .blocklist_domain else .no_match,
        .matched = null,
        .source_id = null,
        // Blocklist provenance, deliberately set on every row: the routes
        // breakdown must never group by it.
        .source_name = "StevenBlack",
        .cname_target = null,
        .safe_search_target = null,
        .route_kind = kind,
        .forward_zone = if (kind == .forward_zone) source else null,
    };
}

test "the type breakdown groups by qtype, keeps the null row and orders it last" {
    var database = try openLog();
    defer database.close();
    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();
    const arena = arena_state.allocator();

    try seed(&database, &.{
        aggRow(0, "192.0.2.10", 1, .upstream, "9.9.9.9"),
        aggRow(1, "192.0.2.10", 1, .upstream, "9.9.9.9"),
        aggRow(2, "192.0.2.10", 1, .upstream, "9.9.9.9"),
        aggRow(3, "192.0.2.10", 28, .upstream, "9.9.9.9"),
        aggRow(4, "192.0.2.10", 28, .upstream, "9.9.9.9"),
        // Ties with qtype 28, so the tie-break puts the lower code first.
        aggRow(5, "192.0.2.10", 16, .upstream, "9.9.9.9"),
        aggRow(6, "192.0.2.10", 16, .upstream, "9.9.9.9"),
        // A row with no type at all: its own group, never a dropped row.
        aggRow(7, "192.0.2.10", null, .upstream, "9.9.9.9"),
        aggRow(8, "192.0.2.10", null, .upstream, "9.9.9.9"),
        // Outside the window.
        aggRow(-1, "192.0.2.10", 255, .upstream, "9.9.9.9"),
    });

    const rows = (try statsTypes(&database, arena, agg_since, agg_until)).items;
    try testing.expectEqual(@as(usize, 4), rows.len);
    try testing.expectEqual(@as(?u16, 1), rows[0].qtype);
    try testing.expectEqual(@as(u64, 3), rows[0].count);
    try testing.expectEqual(@as(?u16, 16), rows[1].qtype);
    try testing.expectEqual(@as(?u16, 28), rows[2].qtype);
    try testing.expectEqual(@as(?u16, null), rows[3].qtype);
    try testing.expectEqual(@as(u64, 2), rows[3].count);
}

test "an empty window has no type rows at all" {
    var database = try openLog();
    defer database.close();
    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();

    const rows = (try statsTypes(&database, arena_state.allocator(), agg_since, agg_until)).items;
    try testing.expectEqual(@as(usize, 0), rows.len);
}

test "the route breakdown keys on the answering resolver, not on blocklist provenance" {
    var database = try openLog();
    defer database.close();
    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();
    const arena = arena_state.allocator();

    try seed(&database, &.{
        aggRow(0, "192.0.2.10", 1, .upstream, "https://a.example/dns-query"),
        aggRow(1, "192.0.2.10", 1, .upstream, "https://a.example/dns-query"),
        aggRow(2, "192.0.2.10", 1, .upstream, "https://b.example/dns-query"),
        // An upstream row whose resolver the log did not record: its own group.
        aggRow(3, "192.0.2.10", 1, .upstream, null),
        aggRow(4, "192.0.2.10", 1, .forward_zone, "lan"),
        aggRow(5, "192.0.2.10", 1, .blocked, null),
        aggRow(6, "192.0.2.10", 1, .cache, null),
        aggRow(7, "192.0.2.10", 1, .local, null),
        aggRow(8, "192.0.2.10", 1, .rejected, null),
    });

    const rows = (try statsRoutes(&database, arena, agg_since, agg_until)).items;
    // Two upstreams, one null-source upstream, one forward zone and four
    // source-less kinds. Every row carries the same `source_name`, so a
    // breakdown that grouped by it would collapse to one row.
    try testing.expectEqual(@as(usize, 8), rows.len);

    try testing.expectEqual(provenance.RouteKind.upstream, rows[0].route);
    try testing.expectEqualStrings("https://a.example/dns-query", rows[0].source.?);
    try testing.expectEqual(@as(u64, 2), rows[0].count);

    // The seven remaining rows all count 1, so the tie-break orders them:
    // route ascending, then source ascending with nulls last.
    for (rows[1..]) |row| try testing.expectEqual(@as(u64, 1), row.count);
    try testing.expectEqual(provenance.RouteKind.blocked, rows[1].route);
    try testing.expectEqual(@as(?[]const u8, null), rows[1].source);
    try testing.expectEqual(provenance.RouteKind.cache, rows[2].route);
    try testing.expectEqual(provenance.RouteKind.forward_zone, rows[3].route);
    try testing.expectEqualStrings("lan", rows[3].source.?);
    try testing.expectEqual(provenance.RouteKind.local, rows[4].route);
    try testing.expectEqual(provenance.RouteKind.rejected, rows[5].route);
    try testing.expectEqual(provenance.RouteKind.upstream, rows[6].route);
    try testing.expectEqualStrings("https://b.example/dns-query", rows[6].source.?);
    try testing.expectEqual(provenance.RouteKind.upstream, rows[7].route);
    try testing.expectEqual(@as(?[]const u8, null), rows[7].source);
}

test "an empty window has no route rows at all" {
    var database = try openLog();
    defer database.close();
    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();

    const rows = (try statsRoutes(&database, arena_state.allocator(), agg_since, agg_until)).items;
    try testing.expectEqual(@as(usize, 0), rows.len);
}

test "an empty window still has a zero-filled other series and no named clients" {
    var database = try openLog();
    defer database.close();
    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();

    const result = try statsClients(&database, arena_state.allocator(), agg_since, agg_width, agg_buckets);
    try testing.expectEqual(@as(usize, 0), result.clients.len);
    try testing.expectEqual(@as(usize, agg_buckets), result.other.len);
    for (result.other) |count| try testing.expectEqual(@as(u64, 0), count);
}

test "client series are bucket-aligned, zero-filled and ranked by in-window total" {
    var database = try openLog();
    defer database.close();
    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();

    try seed(&database, &.{
        aggRow(0, "192.0.2.20", 1, .upstream, "9.9.9.9"),
        aggRow(1, "192.0.2.20", 1, .upstream, "9.9.9.9"),
        aggRow(agg_width * 3, "192.0.2.20", 1, .upstream, "9.9.9.9"),
        aggRow(agg_width * 3, "192.0.2.10", 1, .upstream, "9.9.9.9"),
        // Outside the window on both sides.
        aggRow(-1, "192.0.2.20", 1, .upstream, "9.9.9.9"),
        aggRow(agg_width * agg_buckets, "192.0.2.10", 1, .upstream, "9.9.9.9"),
    });

    const result = try statsClients(&database, arena_state.allocator(), agg_since, agg_width, agg_buckets);
    try testing.expectEqual(@as(usize, 2), result.clients.len);
    try testing.expectEqualStrings("192.0.2.20", result.clients[0].client);
    try testing.expectEqualStrings("192.0.2.10", result.clients[1].client);

    for (result.clients) |series| try testing.expectEqual(@as(usize, agg_buckets), series.buckets.len);
    try testing.expectEqual(@as(u64, 2), result.clients[0].buckets[0]);
    try testing.expectEqual(@as(u64, 1), result.clients[0].buckets[3]);
    try testing.expectEqual(@as(u64, 0), result.clients[0].buckets[9]);
    try testing.expectEqual(@as(u64, 1), result.clients[1].buckets[3]);
    for (result.other) |count| try testing.expectEqual(@as(u64, 0), count);
}

test "the ninth client folds into other and the cut is the same on every read" {
    var database = try openLog();
    defer database.close();
    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();
    const arena = arena_state.allocator();

    // Nine clients, each with one more query than the next, so the ranking is
    // total and the ninth is unambiguously the one that folds.
    var address: [16]u8 = undefined;
    var index: u32 = 0;
    while (index < 9) : (index += 1) {
        const client = try std.fmt.bufPrint(&address, "192.0.2.{d}", .{100 + index});
        var repeat: u32 = 0;
        while (repeat <= index) : (repeat += 1) {
            try seed(&database, &.{aggRow(@intCast(repeat), client, 1, .upstream, "9.9.9.9")});
        }
    }

    const result = try statsClients(&database, arena, agg_since, agg_width, agg_buckets);
    try testing.expectEqual(@as(usize, max_client_series), result.clients.len);
    // The busiest is 192.0.2.108 with nine rows; the lone folded client is
    // 192.0.2.100 with one.
    try testing.expectEqualStrings("192.0.2.108", result.clients[0].client);
    for (result.clients) |series| {
        try testing.expect(!std.mem.eql(u8, "192.0.2.100", series.client));
    }

    var other_total: u64 = 0;
    for (result.other) |count| other_total += count;
    try testing.expectEqual(@as(u64, 1), other_total);
}

test "the three breakdowns conserve the window's total" {
    var database = try openLog();
    defer database.close();
    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();
    const arena = arena_state.allocator();

    // A matrix that exercises every branch the aggregations partition on: null
    // and non-null qtypes, every route kind, a null resolver identity, and ten
    // clients so the top-eight cut has a residual to carry.
    const kinds = [_]provenance.RouteKind{ .blocked, .local, .forward_zone, .upstream, .cache, .rejected };
    var address: [16]u8 = undefined;
    var index: u32 = 0;
    while (index < 40) : (index += 1) {
        const client = try std.fmt.bufPrint(&address, "192.0.2.{d}", .{100 + index % 10});
        const kind = kinds[index % kinds.len];
        try seed(&database, &.{aggRow(
            @intCast(index % (agg_width * agg_buckets)),
            client,
            if (index % 7 == 0) null else @intCast(1 + index % 3),
            kind,
            if (index % 11 == 0) null else "9.9.9.9",
        )});
    }

    const totals = try statsTotals(&database, agg_since, agg_until);
    try testing.expect(totals.queries > 0);

    var typed: u64 = 0;
    for ((try statsTypes(&database, arena, agg_since, agg_until)).items) |row| typed += row.count;
    try testing.expectEqual(totals.queries, typed);

    var routed: u64 = 0;
    for ((try statsRoutes(&database, arena, agg_since, agg_until)).items) |row| routed += row.count;
    try testing.expectEqual(totals.queries, routed);

    var buckets: [agg_buckets]Bucket = undefined;
    _ = try timeseries(&database, agg_since, agg_width, &buckets);
    const clients = try statsClients(&database, arena, agg_since, agg_width, agg_buckets);
    // Per bucket, not just in total: a series misaligned by one bucket would
    // still sum correctly over the window.
    for (buckets, 0..) |bucket, at| {
        var summed: u64 = clients.other[at];
        for (clients.clients) |series| summed += series.buckets[at];
        try testing.expectEqual(bucket.queries, summed);
    }
}

test "the aggregations pass a redacted client through as the log stored it" {
    var database = try openLog();
    defer database.close();
    var arena_state: std.heap.ArenaAllocator = .init(testing.allocator);
    defer arena_state.deinit();

    // `hide_client_ips` is applied by the logger before the row is written, so
    // the read path has nothing to transform — and must not invent anything
    // either. The marker is the client's whole identity here.
    try seed(&database, &.{
        aggRow(0, logger.hidden_marker, 1, .upstream, "9.9.9.9"),
        aggRow(1, logger.hidden_marker, 1, .upstream, "9.9.9.9"),
    });

    const result = try statsClients(&database, arena_state.allocator(), agg_since, agg_width, agg_buckets);
    try testing.expectEqual(@as(usize, 1), result.clients.len);
    try testing.expectEqualStrings(logger.hidden_marker, result.clients[0].client);
    try testing.expectEqual(@as(u64, 2), result.clients[0].buckets[0]);
}

test "a prune committed mid-read is invisible to the reader's transaction" {
    var tmp = testing.tmpDir(.{});
    defer tmp.cleanup();

    var path_buf: [tmp_prefix.len + sub_path_len + 32]u8 = undefined;
    const path = try std.fmt.bufPrintZ(&path_buf, "{s}{s}/querylog.db", .{ tmp_prefix, &tmp.sub_path });

    var reader = try db.Db.open(path, .{ .mode = .read_write_create });
    defer reader.close();
    try db.applyPragmas(&reader, .{});
    try reader.exec(querylog_schema.ddl);
    // The seeded watermark is the file's creation second, which is now; the
    // fixtures below are in 2023, so the prune's cutoff would never advance it.
    try reader.exec(
        \\UPDATE querylog_meta SET created_at = 1700000000, available_since = 1700000000 WHERE id = 1
    );
    try seed(&reader, &.{
        plainRow(agg_since, "a.example"),
        plainRow(agg_since + 1, "b.example"),
        plainRow(agg_since + 300, "c.example"),
    });

    // A second connection to the same file, as retention has in production.
    var pruner = try db.Db.open(path, .{ .mode = .read_write_create });
    defer pruner.close();
    try db.applyPragmas(&pruner, .{});

    var tx = try db.ReadTx.begin(&reader);
    const before = try statsTotals(&reader, agg_since, agg_since + 1000);
    const watermark_before = try availableSince(&reader);
    try testing.expectEqual(@as(u64, 3), before.queries);

    // The prune commits while the reader's transaction is open.
    const pruned = try pruneOlderThan(&pruner, agg_since + 200);
    try testing.expectEqual(@as(i64, 2), pruned.deleted);
    try testing.expect(pruned.available_since > watermark_before);

    // Neither half of the answer moved: the rows the reader would report and
    // the watermark it would tag them with still describe one state.
    const during = try statsTotals(&reader, agg_since, agg_since + 1000);
    try testing.expectEqual(before.queries, during.queries);
    try testing.expectEqual(watermark_before, try availableSince(&reader));
    try tx.commit();

    // The next response sees the prune — both halves of it.
    const after = try statsTotals(&reader, agg_since, agg_since + 1000);
    try testing.expectEqual(@as(u64, 1), after.queries);
    try testing.expectEqual(pruned.available_since, try availableSince(&reader));
}