queue-runner: add prom metrics to allow detecting internal bottlenecks

By looking at the ratio of running vs. waiting for the dispatcher and
the queue monitor, we should get better visibility into what hydra is
currently bottlenecked on.

There are other side effects we can try to measure to get to the same
result, but having a simple way doesn't cost us much.
This commit is contained in:
Pierre Bourdon
2024-04-20 16:48:03 +02:00
committed by John Ericson
parent 1e2d3211d9
commit efcf6815d9
4 changed files with 54 additions and 3 deletions

View File

@@ -42,12 +42,19 @@ void State::queueMonitorLoop(Connection & conn)
bool quit = false;
while (!quit) {
auto t_before_work = std::chrono::steady_clock::now();
localStore->clearPathInfoCache();
bool done = getQueuedBuilds(conn, destStore, lastBuildId);
if (buildOne && buildOneDone) quit = true;
auto t_after_work = std::chrono::steady_clock::now();
prom.queue_monitor_time_spent_running.Increment(
std::chrono::duration_cast<std::chrono::microseconds>(t_after_work - t_before_work).count());
/* Sleep until we get notification from the database about an
event. */
if (done && !quit) {
@@ -72,6 +79,10 @@ void State::queueMonitorLoop(Connection & conn)
printMsg(lvlTalkative, "got notification: jobset shares changed");
processJobsetSharesChange(conn);
}
auto t_after_sleep = std::chrono::steady_clock::now();
prom.queue_monitor_time_spent_waiting.Increment(
std::chrono::duration_cast<std::chrono::microseconds>(t_after_sleep - t_after_work).count());
}
exit(0);