logger tests: join every writer future before its database closes, fix the gate-open race

src/storage/logger.zig

@@ -1334,6 +1334,13 @@ test "shutdown writes the batch the writer holds and the rest of the queue" {
         &database,
         @as(?*disk_monitor.Monitor, null),
     });
+    // Declared after the `database.close` defer so LIFO stops the writer first:
+    // an early return anywhere below would otherwise close the handle under a
+    // live writer and leave `Threaded.deinit` joining a task nothing ends.
+    defer {
+        logger.shutdown(io);
+        future.await(io) catch {};
+    }
 
     var names: [250][32]u8 = undefined;
     for (&names, 0..) |*name, i| {
@@ -1420,6 +1427,12 @@ test "the writer holds an entry for the length of the flush interval" {
         &database,
         @as(?*disk_monitor.Monitor, null),
     });
+    // See the note in "shutdown writes the batch the writer holds": this must
+    // run before the deferred `database.close`.
+    defer {
+        logger.shutdown(io);
+        future.await(io) catch {};
+    }
 
     logger.log(io, sampleEntry(1, "only.example"));
 
@@ -1466,6 +1479,12 @@ test "a full batch flushes without waiting for the interval" {
         &database,
         @as(?*disk_monitor.Monitor, null),
     });
+    // See the note in "shutdown writes the batch the writer holds": this must
+    // run before the deferred `database.close`.
+    defer {
+        logger.shutdown(io);
+        future.await(io) catch {};
+    }
 
     for (0..150) |i| logger.log(io, sampleEntry(@intCast(i), "burst.example"));
 
@@ -1513,6 +1532,12 @@ test "a gated flush holds the batch until the disk recovers" {
         @as([]const Entry, &entries),
         @as(?*disk_monitor.Monitor, &monitor),
     });
+    // This task is `flush`, not `runWriter`: no queue shutdown can release it,
+    // so only cancellation ends it on an early return. Declared after the
+    // `writer.deinit`/`database.close` defers so LIFO runs it first.
+    defer {
+        _ = future.cancel(io) catch {};
+    }
 
     const poll: std.Io.Clock.Duration = .{ .raw = .fromMilliseconds(5), .clock = .awake };
     var waited: usize = 0;
@@ -1634,6 +1659,12 @@ test "the gating episode opens on the gate, turns losing on a drop, and clears o
         &database,
         @as(?*disk_monitor.Monitor, &monitor),
     });
+    // See the note in "shutdown writes the batch the writer holds": this must
+    // run before the deferred `database.close`.
+    defer {
+        logger.shutdown(io);
+        future.await(io) catch {};
+    }
 
     const poll: std.Io.Clock.Duration = .{ .raw = .fromMilliseconds(5), .clock = .awake };
     var waited: usize = 0;
@@ -1657,11 +1688,14 @@ test "the gating episode opens on the gate, turns losing on a drop, and clears o
     // The disk recovers: the held batch goes out and the episode ends.
     monitor.state_raw.store(@intFromEnum(disk_monitor.State.ok), .monotonic);
     waited = 0;
-    while (logger.gateEpisode() != .open) : (waited += 1) {
+    // The write is the completion condition, not the episode: `flush` reopens
+    // the gate before it calls `writeBatch`, so a poll on the episode alone
+    // returns while the row is still in flight and `rows_written` is still 0.
+    while (logger.rows_written.load(.monotonic) == 0) : (waited += 1) {
         try testing.expect(waited < 400);
         try poll.sleep(io);
     }
-    try testing.expect(logger.rows_written.load(.monotonic) > 0);
+    try testing.expectEqual(GateEpisode.open, logger.gateEpisode());
     // The count keeps the history the state does not.
     try testing.expect(logger.queries_dropped.load(.monotonic) > 0);
 
@@ -1736,6 +1770,13 @@ test "a gate that closes mid-enqueue still gets the discard that follows it" {
     // The producer enters `enqueue` with the gate open — which is the reading a
     // sample taken before the loop would keep for the rest of the call.
     var producer = try io.concurrent(logOne, .{ &logger, io });
+    // The producer parks inside `enqueue` and only `release` frees it: an early
+    // return below would otherwise leave `Threaded.deinit` joining it forever.
+    defer {
+        discard_stall.armed = false;
+        discard_stall.release.set(io);
+        producer.await(io);
+    }
     discard_stall.parked.waitUncancelable(io);
 
     // The producer is parked with the row it will evict still on the queue, so
@@ -1799,6 +1840,12 @@ test "a canceled writer counts the batch it was holding" {
         &database,
         @as(?*disk_monitor.Monitor, &monitor),
     });
+    // Cancellation is this case's subject, so the guard is the same operation
+    // the body performs; declared after the `database.close` defer so LIFO ends
+    // the writer before the handle goes away.
+    defer {
+        _ = future.cancel(io) catch {};
+    }
 
     const poll: std.Io.Clock.Duration = .{ .raw = .fromMilliseconds(5), .clock = .awake };
     var waited: usize = 0;
@@ -1842,6 +1889,12 @@ test "a disk-gated writer drops what it holds at shutdown instead of hanging" {
         &database,
         @as(?*disk_monitor.Monitor, &monitor),
     });
+    // See the note in "shutdown writes the batch the writer holds": this must
+    // run before the deferred `fx.deinit` and `database.close`.
+    defer {
+        logger.shutdown(io);
+        future.await(io) catch {};
+    }
 
     const poll: std.Io.Clock.Duration = .{ .raw = .fromMilliseconds(5), .clock = .awake };
     var waited: usize = 0;

src/storage/phase6_integration_test.zig

@@ -211,6 +211,13 @@ test "S8 case 1: the logger writes a real querylog.db end to end" {
         log_db.database(),
         @as(?*disk_monitor.Monitor, null),
     });
+    // Declared after the `log_db.deinit` defer so LIFO stops the writer first:
+    // an early return below would otherwise close the database under a live
+    // writer and leave the never-closed queue parking it forever.
+    defer {
+        query_log.shutdown(io);
+        future.await(io) catch {};
+    }
 
     var name_buf: [32]u8 = undefined;
     for (0..250) |i| {
@@ -252,6 +259,13 @@ test "S8 case 2: a single entry reaches the file once the flush interval passes"
         log_db.database(),
         @as(?*disk_monitor.Monitor, null),
     });
+    // Declared after the `log_db.deinit` defer so LIFO stops the writer first:
+    // an early return below would otherwise close the database under a live
+    // writer and leave the never-closed queue parking it forever.
+    defer {
+        query_log.shutdown(io);
+        future.await(io) catch {};
+    }
 
     query_log.log(io, entryAt(1, "only.example"));
 
@@ -300,6 +314,13 @@ test "S8 case 3: a full queue drops the oldest entries and the newest survive" {
         log_db.database(),
         @as(?*disk_monitor.Monitor, &monitor),
     });
+    // Declared after the `log_db.deinit` defer so LIFO stops the writer first:
+    // an early return below would otherwise close the database under a live
+    // writer and leave the never-closed queue parking it forever.
+    defer {
+        query_log.shutdown(io);
+        future.await(io) catch {};
+    }
 
     try awaitCount(&query_log.batches_gated, 1, 200);
     try testing.expectEqual(@as(i64, 0), try queries_repo.countRows(log_db.database()));
@@ -341,6 +362,13 @@ test "S8 case 4: the privacy transforms reach the stored rows" {
         log_db.database(),
         @as(?*disk_monitor.Monitor, null),
     });
+    // Declared after the `log_db.deinit` defer so LIFO stops the writer first:
+    // an early return below would otherwise close the database under a live
+    // writer and leave the never-closed queue parking it forever.
+    defer {
+        query_log.shutdown(io);
+        future.await(io) catch {};
+    }
 
     var name_buf: [32]u8 = undefined;
     for (0..20) |i| {
@@ -433,6 +461,13 @@ test "S8 case 6: a critical disk gates the flushes and recovery releases them" {
         log_db.database(),
         @as(?*disk_monitor.Monitor, &monitor),
     });
+    // Declared after the `log_db.deinit` defer so LIFO stops the writer first:
+    // an early return below would otherwise close the database under a live
+    // writer and leave the never-closed queue parking it forever.
+    defer {
+        query_log.shutdown(io);
+        future.await(io) catch {};
+    }
 
     try awaitCount(&query_log.batches_gated, 1, 200);
     try testing.expectEqual(@as(u64, 0), query_log.rows_written.load(.monotonic));