Add WithContext methods to the Fluent struct including handling of context deadlines #86
Conversation
…ntext deadlines Signed-off-by: Arno Geurts <[email protected]>
arnogeurts
force-pushed
the
add-with-context-methods
branch
from
2b2330c to
5e1b54c
Compare
|
Ah, I need time to review this change. I'll take time for this later, but please ping me if there will be no further comments in weeks. |
| default: | ||
| return fmt.Errorf("fluent#appendBuffer: Buffer full, limit %v", f.Config.BufferLimit) | ||
| } | ||
| return nil | ||
| } | ||
|
|
||
| // appendBufferWithFeedback appends data to buffer and waits for the result |
There was a problem hiding this comment.
appendBufferWithFeedback did you update the name after writing this comment?
There was a problem hiding this comment.
I did, will change the comment :)
| case f.pending <- bufferInput{msg: msg, result: result}: | ||
| // don't do anything | ||
| case <-msg.ctx.Done(): | ||
| // because the result channel is not used, it can safely be returned to the sync pool. |
There was a problem hiding this comment.
I love this kind of comment 👍
| @@ -386,7 +444,11 @@ func (f *Fluent) run() { | |||
| emitEventDrainMsg.Do(func() { fmt.Fprintf(os.Stderr, "[%s] Discarding queued events...\n", time.Now().Format(time.RFC3339)) }) | |||
| continue | |||
| } | |||
| err := f.write(entry) | |||
| err := f.write(entry.msg) | |||
| if entry.result != nil { | |||
There was a problem hiding this comment.
Are there any cases to get nil on entry.result?
There was a problem hiding this comment.
Yes, so in this change the sync flow and the async flow get very similar. Both write to the pending channel and have a separate go-routine taking entries of that channel and writing the data to the TCP connection.
In case of the async flow no feedback is required, so there will be no result channel in the entry. As you can see in appendBuffer an entry is written to the channel without a result channel.
In case of the sync flow, a result channel is added to the entry, so the feedback can be returned to the caller. And as you can see appendBufferBlocking puts the result channel in the entry and waits on the feedback (hence the "blocking").
| f.wg.Add(1) | ||
| go f.run() |
There was a problem hiding this comment.
Those should be only for Async, right?
There was a problem hiding this comment.
|
Sorry, I couldn't take the time for further review. Let me try in this week or the next week... |
| waitDuration := time.Duration(waitTime) * time.Millisecond | ||
| if deadline, hasDeadLine := msg.ctx.Deadline(); hasDeadLine && deadline.Before(time.Now().Add(waitDuration)) { | ||
| // the context deadline is within the wait time, so after the sleep the deadline will have been | ||
| // exceeded. It is a waste of time to wait on that. |
There was a problem hiding this comment.
Shouldn't it wait until the deadline?
This code is to return DeadlineExceeded when now + waitDuration > deadline but the DeadlineExceeded should occur when the current time is after the deadline.
There was a problem hiding this comment.
If deadline.Before(time.Now()), it should return DeadlineExceeded immediately.
If deadline.Before(time.Now().Add(waitDuration), it should sleep until the deadline.
MaxRetryWait may be configured as minutes or hours, so the current code could cause unexpectedly early return of errors.
| data []byte | ||
| ack string | ||
| } | ||
|
|
||
| type bufferInput struct { | ||
| msg *msgToSend | ||
| result chan<- error |
There was a problem hiding this comment.
This should be syncWriteResult or a similar one to show it's only about sync-write.
| wg sync.WaitGroup | ||
| resultPool sync.Pool |
There was a problem hiding this comment.
Want a different name for this too, like result above. The current name looks so confusing to me.
| @@ -78,17 +79,24 @@ func NewErrUnknownNetwork(network string) error { | |||
| } | |||
|
|
|||
| type msgToSend struct { | |||
| ctx context.Context | |||
There was a problem hiding this comment.
This ctx is referred only in f.write(), and its argument is only the msg. Putting contexts into msgToSend only for passing it to f.write() looks too much to me (because msgToSend is a message, not any other internal structure!)
How about containing ctx in bufferInput? (And add an argument ctx to f.write()?)
It's a structure of msg and internal objects, so it seems a better struct to have the context.
There was a problem hiding this comment.
If we do so, EncodeDataWithContext is not needed anymore, right?
That function looks only for putting contexts to msgToSend structure.
| }) | ||
| } | ||
| } | ||
|
|
There was a problem hiding this comment.
Can you add another test case for the context timeout, via sync and async logger?
Fixes #85
This pull requests adds PostWithContext, PostWithTimeAndContext, EncodeAndPostDataWithContext, EncodeDataWithContext to the Fluent struct. The context deadlines and cancellations are handled at different levels in the code to make sure no unnecessary work is done.