x/build/maintner/maintnerd: /logs endpoint fails to serve 304 successfully during long poll, ends up serving 502 instead #53569
Labels
Builders
x/build issues (builders, bots, dashboards)
FrozenDueToAge
NeedsFix
The path to resolution is known, but the work has not been done.
Projects
Milestone
The /logs endpoint supports long polling via
waitsizenot
parameter. To avoid being flagged as a timeout, it tries to serve a 304 in just under a minute (55 seconds specifically):(Source: x/build/maintner/maintnerd/gcslog/gcslog.go#243-253.)
I think this worked fine before, but in practice it seems not to work at all now. Instead it always serves a 502 after 60 seconds:
Can be reproduced during quiet times (i.e., need 60 seconds+ of inactivity) by manually computing the current corpus size, or by running maintwatch.
Maybe 55 seconds has become too short because of new internal timeouts on the load balancing side. Can trivially try something like 45-50 or 20-25 seconds.
This isn't a big deal for functionality because most callers retry both 502 no differently than a correctly served 304, but adds some noise and inefficiency.
The text was updated successfully, but these errors were encountered: