✨ retry connecting the cloudevents client #280
Conversation
|
/assign @qiujian16 |
|
/hold |
skeeey
changed the title
skeeey
force-pushed
the
cloudevents-connecterr
branch
from
344c2bf
to
89b4d02
Compare
skeeey
changed the title
skeeey
force-pushed
the
cloudevents-connecterr
branch
2 times, most recently
from
2ac7fb0
to
f27e8ba
Compare
skeeey
changed the title
|
/unhold |
cloudevents/generic/baseclient.go
Outdated
| // TODO enhance the cloudevents SKD to avoid wrapping the error type, then we can only handle the | ||
| // net connection errors here | ||
| klog.Errorf("failed to reconnect, %v", err) | ||
| <-connBackoffManager.Backoff().C() |
There was a problem hiding this comment.
if it goes into this branch, will it hang on the select in the next loop?
There was a problem hiding this comment.
the select in the next loop need to wait the backoff, the backoff is a blocking action
cloudevents/generic/baseclient.go
Outdated
| if receiverCancel != nil { | ||
| receiverCancel() | ||
| } | ||
| retrying = true |
There was a problem hiding this comment.
can we just set c.cloudEventsClient = nil here?
There was a problem hiding this comment.
the cloudEventsClient is also used by send method, if we set it to nil here, the send may meet the nil pointer, so I did not set it to nil, what's your suggestion?
There was a problem hiding this comment.
even though, the send will fail right? I think there might need a rwlock in send otherwise there is always a change that the two threads race.
There was a problem hiding this comment.
I mean after backoff it will go to select in the next loop while client is not instantiated, and it will not have a chance to receive message from c.cloudEventsErrChan
There was a problem hiding this comment.
yes, the send will be fail, we maybe need a lock here for the client
the c.cloudEventsErrChan will be only send by a running client, so if the client is retrying, the select will not be blocked the next retry
skeeey
force-pushed
the
cloudevents-connecterr
branch
2 times, most recently
from
b58b1c2
to
d9de5d7
Compare
cloudevents/generic/baseclient.go
Outdated
|
|
||
| for { | ||
| if cloudEventsClient == nil { | ||
| klog.V(4).Infof("reconnecting the cloudevents client") |
There was a problem hiding this comment.
one thing missing is: what if the client just send but not subscribe? Do we alway assume that the client will send and subscribe together?
There was a problem hiding this comment.
hm... I'm not sure, maybe we cannot assume this, in our case, it seems that for our case the send and sub are always together, but as a lib, we don't stop to separate the send and subscribe, user can send in one process, and sub in other process, maybe we need to have one retry mechanism for the client
There was a problem hiding this comment.
separate the reconnect part from the subscription, PTAL
skeeey
force-pushed
the
cloudevents-connecterr
branch
3 times, most recently
from
264e03f
to
4c4356b
Compare
|
|
||
| type receiveFn func(ctx context.Context, evt cloudevents.Event) | ||
|
|
||
| type baseClient struct { |
There was a problem hiding this comment.
should we have a separated Connect(ctx context.Context) in baseClient? so we do not start goroutine during new?
There was a problem hiding this comment.
use connect method instead of new method
Signed-off-by: Wei Liu <liuweixa@redhat.com>
skeeey
force-pushed
the
cloudevents-connecterr
branch
from
4c4356b
to
d80ec98
Compare
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: qiujian16, skeeey The full list of commands accepted by this bot can be found here. The pull request process is described here
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
|
Maybe @yanmxa or @morvencao could also take a look |
| // make sure the current client is the newest | ||
| c.RLock() | ||
| cloudEventsClient = c.cloudEventsClient | ||
| c.RUnlock() |
There was a problem hiding this comment.
do we still need to lock it here? since it is locked at the beginning of the subscribe function
There was a problem hiding this comment.
yes, we need it, the beginning lock is for the subscribe function (it ensure there is only one subscription go routine running), this lock is in a independent go routine (the subscribe lock will be released after this go routine starts)
There was a problem hiding this comment.
emm, right, thanks for the explanation.
|
looks good to me, leave this to @yanmxa |
|
Reference: cloudevents/sdk-go#940 /LGTM |
openshift-merge-robot
merged commit c2b66fa
into
open-cluster-management-io:main
Summary
Retry connecting the cloudevents client when the client has network error