Close and Reopen JIT log files across a checkpoint/restore

[dsouzai]

A common best practice in CRIU mode, at least in OpenJ9, is to close all open files before a checkpoint, and reopen them on restore. The JIT would keep log files such as the vlog and rtLog open across a checkpoint/restore boundary. This PR closes these files on checkpoint and reopens them on restore.

[dsouzai]

A consequence of this change is that when disableSuffixLogs is not specified, a different file is opened. However, this is consistent with what happens for example in the GC.

[dsouzai]

@mpirvu could you please review?

[mpirvu]

Is openNewRTLog(), openNewVlog() and openLogFilesIfNeeded() going to be used anymore?
Also, what about the compilation logs?

[dsouzai]

Also, what about the compilation logs?

I believe these get opened at the start of a compilation and are closed at the end of one.

[dsouzai]

One thing I"m not sure about actually is what happens right now if we specify disableSuffixLogs; I think as the code stands, it'll overwrite the existing file, which is not what we want. So I may need to look into that.

[mpirvu]

Do you to handle disableSuffixLogs in this PR or separately?

[dsouzai]

Do you to handle disableSuffixLogs in this PR or separately?

I figure I should handle it here; I think otherwise, it'll break some of those cmdLineTester tests.

[dsouzai]

Also, what about the compilation logs?

I believe these get opened at the start of a compilation and are closed at the end of one.

I thought this was the case, but it may not be the case... I think we hold on to the file handles so that they can be used by different threads. I'll have to look deeper into this.

[dsouzai]

Also, what about the compilation logs?

I believe these get opened at the start of a compilation and are closed at the end of one.

I thought this was the case, but it may not be the case... I think we hold on to the file handles so that they can be used by different threads. I'll have to look deeper into this.

It is very involved to close and re-open compilation logs across a checkpoint/restore because of the fact that they can be in option sets, shared by different option sets, and shared by different compilation threads. So, I think that's something that should be done in another PR. It may also involve OMR changes.

[dsouzai]

@mpirvu this should be good for review again.

[mpirvu]

LGTM

[mpirvu]

jenkins test sanity all jdk21

[mpirvu]

jenkins test sanity all jdk17

[mpirvu]

Most tests have passed.
Windows builds never finish due to in infra issue.
On AIX there was a failure for jdk_security2_0 due to infra: "Cannot contact p8-java1-ibm07: java.lang.InterruptedException". This is known.

This PR is ready to be merged.