| Author | ![[]](/git/avatar/c50933ca2aa61e0fe2c43d46bb6b59cb_avatar) Andy Green <andy@warmcat.com> 2026-09-25 02:05 UTC | | Committer | Andy Green <andy@warmcat.com> 2026-09-25 02:05 UTC | | Tree | 745ca9a6a3e514889187182ed631d22a7ad0d222 ![]() Raw Patch | | | ![]() server: drop the dead log-driven build_step writer | server: drop the dead log-driven build_step writer
sais_log_to_db() had a second writer of tasks.build_step that matched a
log chunk starting " Step " and took the step number out of it. The
builder's line is ">saib> Step N: [...]", so it has not matched for a
long time; sais_process_rej()'s SAI_TASK_REASON_ACCEPTED handler is the
only thing that advances build_step, and it does so from the db rather
than from log text.
Delete it rather than repair it. Logs reach the db from a 250ms timer,
so a step's line can land after the next step was already accepted, and
rewinding build_step there re-runs a step -- or makes
build_step == build_step_count - 1 come true early, which has the builder
delete the job dir out from under the rest of the build. It also matched
on any child output chunk that happened to begin " Step", so build output
could rewrite the task's step number.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
diff --git a/src/server/s-ws-builder.c b/src/server/s-ws-builder.c
index 284ce37..f72ce04 100644
--- a/src/server/s-ws-builder.c
+++ b/src/server/s-ws-builder.c
@@ -206,11 +206,8 @@ sais_dump_logs_to_db(lws_sorted_usec_list_t *sul)
static void
sais_log_to_db(struct vhd *vhd, sai_log_t *log)
{
- char event_uuid[33], q[256], esc_uuid[129];
sais_logcache_pertask_t *lcpt = NULL;
- sqlite3 *pdb = NULL;
sai_log_t *hlog;
- int step;
if (!log || !log->log)
return;
@@ -313,27 +310,20 @@ sais_log_to_db(struct vhd *vhd, sai_log_t *log)
}
}
- if (log->len < 5 || memcmp(log->log, " Step ", 5))
- return;
-
- step = atoi(&log->log[5]);
-
- sai_task_uuid_to_event_uuid(event_uuid, log->task_uuid);
-
- if (sai_event_db_ensure_open(vhd->context, &vhd->sqlite3_cache,
- vhd->sqlite3_path_lhs, event_uuid, 0, &pdb))
- return;
-
- lws_sql_purify(esc_uuid, log->task_uuid, sizeof(esc_uuid));
-
- lws_snprintf(q, sizeof(q),
- "UPDATE tasks SET build_step=%d WHERE uuid='%s' and run=(select max(run) from tasks where uuid='%s')",
- step, esc_uuid, esc_uuid);
-
- if (sai_sqlite3_statement(pdb, q, "update build_step"))
- lwsl_err("%s: failed to update build_step\n", __func__);
-
- sai_event_db_close(&vhd->sqlite3_cache, &pdb);
+ /*
+ * There used to be a second, log-driven writer of tasks.build_step
+ * here, matching a chunk starting " Step " and taking the step number
+ * out of it. The builder's line is ">saib> Step N: [...]", so it never
+ * matched and has been dead for a long time; sais_process_rej()'s
+ * SAI_TASK_REASON_ACCEPTED handler is the only thing that advances
+ * build_step, and it does so from the DB rather than from log text.
+ *
+ * It isn't worth reviving: logs reach the db from a 250ms timer, so a
+ * step's line can land after the next step was already accepted, and
+ * rewinding build_step there re-runs a step -- or makes
+ * build_step == build_step_count - 1 come true early, which has the
+ * builder delete the job dir out from under the rest of the build.
+ */
}
sai_plat_t *
|