1032851561 commented on issue #8087:
URL: https://github.com/apache/hudi/issues/8087#issuecomment-1479189389
> But for each input split, we only comsume a 2014 records batch a time, the
checkpoint barrier should fetch the checkpoint lock while one batch consumption
is done ?
I think checkpoint lock is not used here. Using mailbox instead of, the mini
batch consumption was packaged as mail to submit to `TaskMailBox`.
`StreamTask#runMailboxLoop` -> `MailboxProcessor#runMailboxLoop` run as below:
```
while (true) {
processMail(localMailbox, false); // 1. process current all
mails (mini batch consumption)
mailboxDefaultAction.runDefaultAction(defaultActionContext); //
2. processElement(MergeOnReadInputSplit) or process barriers
}
```
There are two things will submit a “mini batch consumption” mail to MailBox:
1. `StreamReadOperator#processElement`
2. after `consumeAsMiniBatch`
The first checkpoint barrier is behind 1200 splits. When it is comming,
split_read has consumed many mini batchs(maybe not all split), but it takes a
long time.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
