v1.8: hubble/relay: flush old flows when the buffer drain timeout is reached #13877
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
[ upstream commit 195351a ] Signed-off-by: Robin Hahling <robin.hahling@gw-computing.net>
[ upstream commit 7d56068 ] Draining only 1 flow every time the buffer drain timeout is reached proves to be insufficient for certain cases. The worst case typically happens for a request with historic data, some filters and follow-mode. Using the Hubble CLI, this means requests like this one: hubble observe --last=100 --follow --pod kube-system/coredns-f9fd979d6-66mfg The behavior before this patch for a request like this, assuming 250 flows matching the requests are stored in Hubble instances ring buffers and new flows matching the filter criteria are infrequent, would be to forward 150 flows, then drain 1 flow to forward every time the drain timeout is reached. This means that for a buffer size of 100, it would take at least 100 seconds (with the default buffer drain timeout of 1s) to drain the buffer (unless enough new flows are received in the meantime and fill the buffer again). The new code drains all flows in the queue for which the timestamp is older than `now-bufferDrainTimeout`. In other words, this would be the old behavior: ... QUEUE FULL draining flow 149 QUEUE FULL draining flow 150 TIMEOUT REACHED DRAINED 1 flow TIMEOUT REACHED DRAINED 1 flow TIMEOUT REACHED DRAINED 1 flow ... And this is the new behavior: ... QUEUE FULL draining flow 149 QUEUE FULL draining flow 150 TIMEOUT REACHED DRAINED 100 flow TIMEOUT REACHED DRAINED 0 flow TIMEOUT REACHED DRAINED 0 flow ... If new flows are received but not in enough numbers to fill the buffer within the time window of `bufferDrainTimeout`, the behavior would look something like this: ... QUEUE FULL draining flow 149 QUEUE FULL draining flow 150 TIMEOUT REACHED DRAINED 100 flow TIMEOUT REACHED DRAINED 34 flow TIMEOUT REACHED DRAINED 72 flow ... In effect, this means that a time window of `bufferDrainTimeout` (default to 1s) is considered to sort flows before forwarding them to the client. This dramatically improves the query experience for requests in follow-mode. Suggested-by: Tom Payne <tom@isovalent.com> Signed-off-by: Robin Hahling <robin.hahling@gw-computing.net>
rolinh
added
kind/backports
|
test-backport-1.8 |
maintainer-s-little-helper
bot
added
the
ready-to-merge
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This is a manual backport of #13776 to v1.8 as protobuf's timestamp in version in 1.8 does not have the
AsTime()method which results in compilation failure. This is worked around in this PR by implementing a helper which converts atimestamp.Timestampto atime.Timeobject (this is covered by unit tests).Ref: #13875 cc @tklauser
Once this PR is merged, you can update the PR labels via: