RFR: 8351594: JFR: Rate-limited sampling of Java events [v3]

[Erik Gahlin]

On Wed, 4 Jun 2025 14:32:31 GMT, Alan Bateman <alanb at openjdk.org> wrote:

>> Erik Gahlin has updated the pull request incrementally with one additional commit since the last revision:
>> 
>>   Fix adjust boundary
>
> src/java.base/share/classes/java/net/Socket.java line 970:
> 
>> 968:             long end = SocketReadEvent.timestamp();
>> 969:             long duration = end - start;
>> 970:             if (SocketReadEvent.shouldThrottleCommit(duration, end)) {
> 
> The use sites want to ask if an event should be committed. Does the word "Throttle" need to be in name as I don't think the use cases need to care about this.
> 
> Also,  just to point out that the shouldXXX method is called with the the end timestamp and duration whereas the offset/emit/commit methods are called with the start timestamp and duration. So two long timestamps and a long measure of time, easy to get it wrong somewhere. Maybe not this PR but I think would be clearer at the use sites to use start or end consistently and reduce potential for mistakes.

We need some indication of which events are throttleable and looking at the mirror event may not work in some scenarios.

We need to sample the endTime, because the startTime may be several minutes in the past. We could use commit(startTime, endTime, ...) and calculate the duration inside the method, but it may be confusing because the fields in the event are startTime and duration. We would also need to calculate the duration twice, both for shouldCommit and commit.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/25559#discussion_r2126960704