LogDog: collector cache non-transient errors.

LogDog: collector cache non-transient errors.

Currently, the collector will cache responses from the Coordinator only
on success. Now, non-transient collector errors will also be cached,
meaning we won't hammer Collector with data that is known to be invalid.

This is in anticipation of a higher registration failure rate after a
prefix space expires. Currently, each log bundle for that stream will cause
a new RPC to be issued. After this CL, the non-transient failure state
will be cached.

BUG=

[dnj (Google), 2016-05-11 23:11:26]

dnj@google.com changed reviewers:
+ dnj@google.com, hinoka@chromium.org, iannucci@chromium.org

[dnj (Google), 2016-05-11 23:11:26]

PTAL

[dnj, 2016-05-17 20:41:48]

Description was changed from

==========
LogDog: collector cache non-transient errors.

Currently, the collector will cache responses from the Coordinator only
on success. Now, non-transient collector errors will also be cached,
meaning we won't hammer Collector with data that is known to be invalid.

BUG=
==========

to

==========
LogDog: collector cache non-transient errors.

Currently, the collector will cache responses from the Coordinator only
on success. Now, non-transient collector errors will also be cached,
meaning we won't hammer Collector with data that is known to be invalid.

This is in anticipation of a higher registration failure rate after a
prefix space expires. Currently, each log bundle for that stream will cause
a new RPC to be issued. After this CL, the non-transient failure state
will be cached.

BUG=
==========

[nodir, 2016-05-19 01:01:51]

nodir@chromium.org changed reviewers:
+ nodir@chromium.org

[nodir, 2016-05-19 01:01:51]

lgtm

https://codereview.chromium.org/1971933002/diff/20001/server/internal/logdog/...
File server/internal/logdog/collector/coordinator/cache.go (right):

https://codereview.chromium.org/1971933002/diff/20001/server/internal/logdog/...
server/internal/logdog/collector/coordinator/cache.go:99: st, err :=
c.Coordinator.RegisterStream(ctx, st, desc)
just return it

https://codereview.chromium.org/1971933002/diff/20001/server/internal/logdog/...
server/internal/logdog/collector/coordinator/cache.go:115: // that the next
"update" will re-fetch it.
I am not clear where this purge happens?

[dnj, 2016-07-12 15:22:45]

Description was changed from

==========
LogDog: collector cache non-transient errors.

Currently, the collector will cache responses from the Coordinator only
on success. Now, non-transient collector errors will also be cached,
meaning we won't hammer Collector with data that is known to be invalid.

This is in anticipation of a higher registration failure rate after a
prefix space expires. Currently, each log bundle for that stream will cause
a new RPC to be issued. After this CL, the non-transient failure state
will be cached.

BUG=
==========

to

==========
LogDog: collector cache non-transient errors.

Currently, the collector will cache responses from the Coordinator only
on success. Now, non-transient collector errors will also be cached,
meaning we won't hammer Collector with data that is known to be invalid.

This is in anticipation of a higher registration failure rate after a
prefix space expires. Currently, each log bundle for that stream will cause
a new RPC to be issued. After this CL, the non-transient failure state
will be cached.

BUG=
==========