On Mon, 25 Nov 2024 10:59:35 GMT, Erik Gahlin <egah...@openjdk.org> wrote:
>>> Perhaps this should be made more explicit with a failed field. >>> Alternatively, there could be two events: one for success and one for >>> failure. What is the typical duration of a failed event? If it is above >>> 10-20 ms, two events might not be as useful since all failures will be >>> recorded anyway. >> >> If a connection cannot be established then it might be immediate, 10s of >> milliseconds, maybe 60+ seconds in some cases. A slow down or stall waiting >> for a connection to be established seems a useful event to have recorded. > >> If a connection cannot be established then it might be immediate, 10s of >> milliseconds, maybe 60+ seconds in some cases. A slow down or stall waiting >> for a connection to be established seems a useful event to have recorded. > > If it's immediate, a potential Socket Connection Failure event could overflow > the buffers and we can't have it with threshold = 0s. Otherwise, it might be > interesting to have something like: > > `$ jfr view socket-connection-failures recording.jfr` > > to see a complete list of failures per host/port and then have: > > `$ jfr view slow-socket-connections recording.jfr` > > to find which ones are slow, i.e. more than 10-20 ms. Having a view for connect failures that doesn't require exceptions=all could be useful. Does this mean two events, one an instant event for the failures, the other a duration event for the connections that are slow to establish? ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/21528#discussion_r1856496052
