fix(worker): record real raw token usage per run, not the uncached remainder

task_runs.tokens_in only ever held the API's uncached "input_tokens" field
(off by a factor of ~400,000 on a resumed session), and tokens_out summed
only the last result event instead of the whole session. TaskRunner now
reads each run's cache-read/cache-write/input/output totals from the
session transcript via a new ITranscriptUsageReader.ReadSessionTotalsAsync,
storing the delta against prior runs on the same session so a --resume
doesn't double-count. New task_runs.cache_read_tokens/cache_write_tokens
columns; the Session tab now shows the raw total (what actually counts
against the 5h/7d usage limit) with a breakdown tooltip.
This commit is contained in:
mika kuns
2026-08-05 15:44:20 +02:00
parent 83ea429b8a
commit 7d3d6d7b54
28 changed files with 1318 additions and 20 deletions
@@ -56,7 +56,7 @@ public sealed class QueueServiceSlotGuardTests : IDisposable
var built = TaskStateServiceBuilder.Build(dbFactory);
var state = built.State;
var runner = new TaskRunner(fake, dbFactory, broadcaster, wtManager, argsBuilder, _cfg,
NullLogger<TaskRunner>.Instance, state, new TaskRunTokenRegistry(), new AttachmentStore(), new FakeSessionSkillSeeder());
NullLogger<TaskRunner>.Instance, state, new TaskRunTokenRegistry(), new AttachmentStore(), new FakeSessionSkillSeeder(), new FakeTranscriptUsageReader());
_waker = new QueueWaker();
var picker = new QueuePicker(dbFactory);
var overrideSlot = new OverrideSlotService(dbFactory, runner, NullLogger<OverrideSlotService>.Instance, built.RunCancels);