[3.5] gRPC health server sets serving status to NOT_SERVING on defrag #17914
Conversation
tjungblu
changed the title
gRPC health server sets serving status to NOT_SERVING on defrag Backport from 3.6 in etcd-io#16278 Co-authored-by: Chao Chen <chaochn@amazon.com> Signed-off-by: Thomas Jungblut <tjungblu@redhat.com>
tjungblu
force-pushed
the
backport_35_grpc_defragserv
branch
from
1794125 to
750bc0b
Compare
|
the grpcProxy failure seems to be #16475 which wasn't backported to 3.5 |
|
Click approve too soon: there is a follow up PR #16836 which improves code readability. Could you please also take that one? There is another in flight PR fixing the insecure / secure RPC servers set up #16959.. Although this is not a blocker of this PR since this is protected by an experimental feature flag and still improves the availability, we should document that in the help messages clearly. |
|
Thanks @chaochn47, I'll pick the other PR too and let you know |
Backport from 3.6 in etcd-io#16836 Co-authored-by: Chao Chen <chaochn@amazon.com> Signed-off-by: Thomas Jungblut <tjungblu@redhat.com>
|
@chaochn47 please take a look again with your refactoring applied otherwise @ahrtr and @serathius PTAL |
| @@ -301,6 +301,7 @@ func newConfig() *config { | |||
| fs.DurationVar(&cfg.ec.ExperimentalWarningApplyDuration, "experimental-warning-apply-duration", cfg.ec.ExperimentalWarningApplyDuration, "Time duration after which a warning is generated if request takes more time.") | |||
| fs.BoolVar(&cfg.ec.ExperimentalMemoryMlock, "experimental-memory-mlock", cfg.ec.ExperimentalMemoryMlock, "Enable to enforce etcd pages (in particular bbolt) to stay in RAM.") | |||
| fs.BoolVar(&cfg.ec.ExperimentalTxnModeWriteWithSharedBuffer, "experimental-txn-mode-write-with-shared-buffer", true, "Enable the write transaction to use a shared buffer in its readonly check operations.") | |||
| fs.BoolVar(&cfg.ec.ExperimentalStopGRPCServiceOnDefrag, "experimental-stop-grpc-service-on-defrag", cfg.ec.ExperimentalStopGRPCServiceOnDefrag, "Enable etcd gRPC service to stop serving client requests on defragmentation.") | |||
There was a problem hiding this comment.
Adding experimental prefix and removing it in future is a breaking change as we discussed in #17657. Also see #17657
Are we still happy to add prefix experimental for an flag?
I know this PR just backports changes from main to 3.5, so the suggestion (adding experimental in description only) is actually for the main branch (update main firstly). Do we want to follow the suggestion for now? Or do we have an agreement on #17657 and #17657 before we add any new flags? @fuweid @jmhbnz @serathius @siyuanfoundation @spzala
| fs.BoolVar(&cfg.ec.ExperimentalStopGRPCServiceOnDefrag, "experimental-stop-grpc-service-on-defrag", cfg.ec.ExperimentalStopGRPCServiceOnDefrag, "Enable etcd gRPC service to stop serving client requests on defragmentation.") | |
| fs.BoolVar(&cfg.ec.ExperimentalStopGRPCServiceOnDefrag, "stop-grpc-service-on-defrag", cfg.ec.ExperimentalStopGRPCServiceOnDefrag, "(Experimental) Enable etcd gRPC service to stop serving client requests on defragmentation.") |
There was a problem hiding this comment.
As part of #17657 we will still need to support migration from the old flag naming scheme (with prefix) to the new one. One more flag should not change much as we already have other flags to migrate. So I don't think we need to block on this.
There was a problem hiding this comment.
I'm happy for this to proceed using older approach. We are still working on the KEP for feature flags so it will be a while off and agree with @serathius above we don't want to block this work.
LGTM, thanks @tjungblu !! |
gRPC health server sets serving status to NOT_SERVING on defrag.
Backported from 3.6 in #16278
would be awesome to have this available in 3.5 already, we ran into this recently with a performance test when running etcd with 38 gigs.