fix: reset stale execution state after SIGUSR1 in-process restart (#15195)
Merged via /review-pr -> /prepare-pr -> /merge-pr. Prepared head SHA: 676f9ec45135be0d3471bb0444bc2ac7ce7d5224 Co-authored-by: joeykrug <5925937+joeykrug@users.noreply.github.com> Co-authored-by: gumadeiras <5599352+gumadeiras@users.noreply.github.com> Reviewed-by: @gumadeiras
This commit is contained in:
@@ -29,10 +29,10 @@ type QueueEntry = {
|
||||
type LaneState = {
|
||||
lane: string;
|
||||
queue: QueueEntry[];
|
||||
active: number;
|
||||
activeTaskIds: Set<number>;
|
||||
maxConcurrent: number;
|
||||
draining: boolean;
|
||||
generation: number;
|
||||
};
|
||||
|
||||
const lanes = new Map<string, LaneState>();
|
||||
@@ -46,15 +46,23 @@ function getLaneState(lane: string): LaneState {
|
||||
const created: LaneState = {
|
||||
lane,
|
||||
queue: [],
|
||||
active: 0,
|
||||
activeTaskIds: new Set(),
|
||||
maxConcurrent: 1,
|
||||
draining: false,
|
||||
generation: 0,
|
||||
};
|
||||
lanes.set(lane, created);
|
||||
return created;
|
||||
}
|
||||
|
||||
function completeTask(state: LaneState, taskId: number, taskGeneration: number): boolean {
|
||||
if (taskGeneration !== state.generation) {
|
||||
return false;
|
||||
}
|
||||
state.activeTaskIds.delete(taskId);
|
||||
return true;
|
||||
}
|
||||
|
||||
function drainLane(lane: string) {
|
||||
const state = getLaneState(lane);
|
||||
if (state.draining) {
|
||||
@@ -63,7 +71,7 @@ function drainLane(lane: string) {
|
||||
state.draining = true;
|
||||
|
||||
const pump = () => {
|
||||
while (state.active < state.maxConcurrent && state.queue.length > 0) {
|
||||
while (state.activeTaskIds.size < state.maxConcurrent && state.queue.length > 0) {
|
||||
const entry = state.queue.shift() as QueueEntry;
|
||||
const waitedMs = Date.now() - entry.enqueuedAt;
|
||||
if (waitedMs >= entry.warnAfterMs) {
|
||||
@@ -74,29 +82,31 @@ function drainLane(lane: string) {
|
||||
}
|
||||
logLaneDequeue(lane, waitedMs, state.queue.length);
|
||||
const taskId = nextTaskId++;
|
||||
state.active += 1;
|
||||
const taskGeneration = state.generation;
|
||||
state.activeTaskIds.add(taskId);
|
||||
void (async () => {
|
||||
const startTime = Date.now();
|
||||
try {
|
||||
const result = await entry.task();
|
||||
state.active -= 1;
|
||||
state.activeTaskIds.delete(taskId);
|
||||
diag.debug(
|
||||
`lane task done: lane=${lane} durationMs=${Date.now() - startTime} active=${state.active} queued=${state.queue.length}`,
|
||||
);
|
||||
pump();
|
||||
const completedCurrentGeneration = completeTask(state, taskId, taskGeneration);
|
||||
if (completedCurrentGeneration) {
|
||||
diag.debug(
|
||||
`lane task done: lane=${lane} durationMs=${Date.now() - startTime} active=${state.activeTaskIds.size} queued=${state.queue.length}`,
|
||||
);
|
||||
pump();
|
||||
}
|
||||
entry.resolve(result);
|
||||
} catch (err) {
|
||||
state.active -= 1;
|
||||
state.activeTaskIds.delete(taskId);
|
||||
const completedCurrentGeneration = completeTask(state, taskId, taskGeneration);
|
||||
const isProbeLane = lane.startsWith("auth-probe:") || lane.startsWith("session:probe-");
|
||||
if (!isProbeLane) {
|
||||
diag.error(
|
||||
`lane task error: lane=${lane} durationMs=${Date.now() - startTime} error="${String(err)}"`,
|
||||
);
|
||||
}
|
||||
pump();
|
||||
if (completedCurrentGeneration) {
|
||||
pump();
|
||||
}
|
||||
entry.reject(err);
|
||||
}
|
||||
})();
|
||||
@@ -134,7 +144,7 @@ export function enqueueCommandInLane<T>(
|
||||
warnAfterMs,
|
||||
onWait: opts?.onWait,
|
||||
});
|
||||
logLaneEnqueue(cleaned, state.queue.length + state.active);
|
||||
logLaneEnqueue(cleaned, state.queue.length + state.activeTaskIds.size);
|
||||
drainLane(cleaned);
|
||||
});
|
||||
}
|
||||
@@ -155,13 +165,13 @@ export function getQueueSize(lane: string = CommandLane.Main) {
|
||||
if (!state) {
|
||||
return 0;
|
||||
}
|
||||
return state.queue.length + state.active;
|
||||
return state.queue.length + state.activeTaskIds.size;
|
||||
}
|
||||
|
||||
export function getTotalQueueSize() {
|
||||
let total = 0;
|
||||
for (const s of lanes.values()) {
|
||||
total += s.queue.length + s.active;
|
||||
total += s.queue.length + s.activeTaskIds.size;
|
||||
}
|
||||
return total;
|
||||
}
|
||||
@@ -180,6 +190,36 @@ export function clearCommandLane(lane: string = CommandLane.Main) {
|
||||
return removed;
|
||||
}
|
||||
|
||||
/**
|
||||
* Reset all lane runtime state to idle. Used after SIGUSR1 in-process
|
||||
* restarts where interrupted tasks' finally blocks may not run, leaving
|
||||
* stale active task IDs that permanently block new work from draining.
|
||||
*
|
||||
* Bumps lane generation and clears execution counters so stale completions
|
||||
* from old in-flight tasks are ignored. Queued entries are intentionally
|
||||
* preserved — they represent pending user work that should still execute
|
||||
* after restart.
|
||||
*
|
||||
* After resetting, drains any lanes that still have queued entries so
|
||||
* preserved work is pumped immediately rather than waiting for a future
|
||||
* `enqueueCommandInLane()` call (which may never come).
|
||||
*/
|
||||
export function resetAllLanes(): void {
|
||||
const lanesToDrain: string[] = [];
|
||||
for (const state of lanes.values()) {
|
||||
state.generation += 1;
|
||||
state.activeTaskIds.clear();
|
||||
state.draining = false;
|
||||
if (state.queue.length > 0) {
|
||||
lanesToDrain.push(state.lane);
|
||||
}
|
||||
}
|
||||
// Drain after the full reset pass so all lanes are in a clean state first.
|
||||
for (const lane of lanesToDrain) {
|
||||
drainLane(lane);
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Returns the total number of actively executing tasks across all lanes
|
||||
* (excludes queued-but-not-started entries).
|
||||
@@ -187,7 +227,7 @@ export function clearCommandLane(lane: string = CommandLane.Main) {
|
||||
export function getActiveTaskCount(): number {
|
||||
let total = 0;
|
||||
for (const s of lanes.values()) {
|
||||
total += s.active;
|
||||
total += s.activeTaskIds.size;
|
||||
}
|
||||
return total;
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user