kvserver: apply a limit to outgoing raft msg batching #71132
Conversation
|
This change is |
|
Still need to write a test, which... I feel like that will require me to do a lot of refactoring that I don't want to do (backports), so it will be in a separate commit. |
There was a problem hiding this comment.
Reviewed 1 of 1 files at r1, 1 of 1 files at r2, all commit messages.
Reviewable status:complete! 0 of 0 LGTMs obtained (waiting on @tbg)
pkg/kv/kvserver/raft_transport.go, line 494 at r1 (raw file):
// Pull off as many queued requests as possible. // // TODO(peter): Think about limiting the size of the batch we send.
Feels good to resolve a five-year-old TODO!
pkg/kv/kvserver/raft_transport.go, line 490 at r2 (raw file):
return err case req := <-ch: maxBytes := MaxCommandSize.Get(&t.st.SV)
You may want to note that this doesn't actually have anything to do with limiting the size of a command, we're just applying the same limit to something semi-related. We could even have a separate setting for it, but I think this is reasonable.
There was a problem hiding this comment.
Reviewable status:
pkg/kv/kvserver/raft_transport.go, line 510 at r1 (raw file):
} // Reuse the Requests slice, but zero out the contents to avoid delaying
Nice spot.
pkg/kv/kvserver/raft_transport.go, line 490 at r2 (raw file):
return err case req := <-ch: maxBytes := MaxCommandSize.Get(&t.st.SV)
Should we be accounting for this first req?
pkg/kv/kvserver/raft_transport.go, line 490 at r2 (raw file):
Previously, erikgrinaker (Erik Grinaker) wrote…
You may want to note that this doesn't actually have anything to do with limiting the size of a command, we're just applying the same limit to something semi-related. We could even have a separate setting for it, but I think this is reasonable.
I think a separate setting is wise. We may want to tune these knobs separately and this has a very different meaning than MaxCommandSize. It also has a different meaning than a lot of the per-range limits we pass to etcd/raft because this limit is above the level at which we multiplex multiple ranges onto the same transport.
There was a problem hiding this comment.
Reviewable status:
pkg/kv/kvserver/raft_transport.go, line 490 at r2 (raw file):
Previously, nvanbenschoten (Nathan VanBenschoten) wrote…
I think a separate setting is wise. We may want to tune these knobs separately and this has a very different meaning than
MaxCommandSize. It also has a different meaning than a lot of the per-range limits we pass toetcd/raftbecause this limit is above the level at which we multiplex multiple ranges onto the same transport.
Yeah, I also know of users who run with this setting set very large (several GB), because they got burnt when we accidentally generated commands that were too large. The occasional huge command is probably fine, but shipping requests that large around semi-regularly maybe isn't.
|
Sounds good, will ping this PR once I've addressed these comments after having caught back up from PTO. |
Standard Go error - we were trying to avoid allocations by recycling a slice but weren't zeroing it out before. The occasional long slice that would reference a ton of memory would then effectively keep that large amount of memory alive forever. Touches cockroachdb#71050. Release note: None
tbg
force-pushed
the
fix-giant-raft-msgs
branch
from
9ccf33e
to
3fc9c27
Compare
|
Updated, PTAL |
There was a problem hiding this comment.
Reviewed 1 of 1 files at r3, 1 of 1 files at r4, all commit messages.
Reviewable status:
pkg/kv/kvserver/raft_transport.go, line 59 at r4 (raw file):
// maxRaftOutgoingBatchSize wraps "kv.raft.command.max_batch_size". var maxRaftOutgoingBatchSize = settings.RegisterByteSizeSetting( "kv.raft.command.max_batch_size",
This should be target_batch_size, not max, since we generally overshoot it.
tbg
force-pushed
the
fix-giant-raft-msgs
branch
from
3fc9c27
to
7f9adeb
Compare
tbg
changed the title
kv.raft.command.max_size to outgoing raft msg batching
There was a problem hiding this comment.
Updated. Not sure if you wanted to re-approve this, but if it looks good you might as well bors it for me.
Dismissed @erikgrinaker from 2 discussions.
Reviewable status:
pkg/kv/kvserver/raft_transport.go, line 59 at r4 (raw file):
Previously, erikgrinaker (Erik Grinaker) wrote…
This should be
target_batch_size, notmax, since we generally overshoot it.
Done.
There was a problem hiding this comment.
Reviewed 1 of 1 files at r5, all commit messages.
Reviewable status:
-- commits, line 21 at r5:
nit: s/traget/target/
pkg/kv/kvserver/raft_transport.go, line 505 at r5 (raw file):
return err case req := <-ch: maxBytes := targetRaftOutgoingBatchSize.Get(&t.st.SV) - int64(req.Size())
nit: may want to rename maxBytes as well.
… msg batching In cockroachdb#71050, we saw evidence of very large (2.3+GiB) Raft messages being sent on the stream, which overwhelmed both the sender and the receiver. Raft messages are batched up before sending and so what must have happened here is that a large number of reasonably-sized messages (up to 64MiB in this case due to the max_size setting) were merged into a giant blob. As of this commit, we apply the target-size chunking on the batching step before sending messages as well. Closes cockroachdb#71050. Release note: None
tbg
force-pushed
the
fix-giant-raft-msgs
branch
from
7f9adeb
to
1d959cb
Compare
|
bors r=erikgrinaker |
|
Build succeeded: |
In #71050, we saw evidence of very large (2.3+GiB) Raft messages being
sent on the stream, which overwhelmed both the sender and the receiver.
Raft messages are batched up before sending and so what must have
happened here is that a large number of reasonably-sized messages (up to
64MiB in this case due to the
max_sizesetting) were merged into a giantblob. As of this commit, we apply the
max_sizechunking on the batchingstep before sending messages as well.
Closes #71050.
Release note: None