Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
    Carlos Andres De La Rosa
    @carlosdlr
    if the issue is infra like disk space memory etc. the approach changes
    but the task need to finish
    if not, not make sense this pattern
    Ladislav Thon
    @Ladicek
    we should probably go further than what we do now, I think
    we successfully prevent "more of such problems", but we don't specify how to let the user know
    well
    in case the queue is too big :-)
    is that the case you're thinking of @jbee ? the task enters the bulkhead queue, because there's capacity, but is never allowed to actually run
    when the queue is full, you'll get an exception immediately
    Jan Bernitt
    @jbee
    I was thinking that for some reason some of the computations never complete - such would always block 1 slot of the capacity and if the code allows for such a path it is likely that it will soon or later gobble all slots with such stale executions. that is not a situation of high load - just that some code path cause never completing computations. that would then block the bulkhead completely but from the point of view of the annotated method there is no work at all going on so it kind of fails the point of being a bulkhead
    I might have a way to implement it in a way that is somewhat "self-healing". As I am remembering the actual threads that are computing should a thread trigger a stale computation but then get reused for another one on the same method it could check if it is already considered running in the bulkhead an in that case enter without changing the count
    Andrew Rouse
    @Azquelt
    The assumption inherent in Bulkhead is that a running task consumes some resource, so we want to avoid them building up. This would have to be a situation where the task is not complete but no resource is being consumed.
    Jan Bernitt
    @jbee
    first I can think of are work items submitted to a pool where each item has a timeout. if the user implements this without completing the stage even in case the work item is cancelled it would be such an abandoned computation
    Carlos Andres De La Rosa
    @carlosdlr
    is not abandoned is cancelled
    you already say it
    Jan Bernitt
    @jbee
    yeah, so we would say its a user error that causes this and the fix that the user code has to make sure the computation is either completed or cancelled
    Carlos Andres De La Rosa
    @carlosdlr
    i think we have a misconception in what is a computational process. Input => some process => result
    Andrew Rouse
    @Azquelt
    Yeah
    Carlos Andres De La Rosa
    @carlosdlr
    always is a result
    called in the way that you want
    doesn't matter
    Andrew Rouse
    @Azquelt
    It probably does make sense for implementations to warn the user if a task in the bulkhead appears to have hung, but that's not something that the spec would mandate.
    Jan Bernitt
    @jbee
    FYI: the TCK tests do not detect when a CompletionStagereturning @Bulkhead without @Asychronous wrongly gets the different semantics
    Andrew Rouse
    @Azquelt
    :(
    Emily Jiang
    @Emily-Jiang
    I need to prep for something urgent. Won't be able to join the call today. @Azquelt is off today.
    Carlos Andres De La Rosa
    @carlosdlr
    me also cannot join
    in another meeting now
    Ladislav Thon
    @Ladicek
    do we cancel then? I'd be in favor, I also have urgent work to do :-)
    ... ok, I'm declaring today call cancelled :-)
    Emily Jiang
    @Emily-Jiang
    sure.
    talk to you next week
    Emily Jiang
    @Emily-Jiang
    will be late to join today, probably 15mins
    who will join the call
    Andrew Rouse
    @Azquelt
    Me
    Carlos Andres De La Rosa
    @carlosdlr
    here
    Andrew Rouse
    @Azquelt
    @Emily-Jiang Do you want us to wait for you?
    Emily Jiang
    @Emily-Jiang
    joining. you can start
    Emily Jiang
    @Emily-Jiang
    Do we have agenda to discuss today?
    Emily Jiang
    @Emily-Jiang
    I'll be joining 10mins late if you guys decide to have a call. I don't have anything to discuss but I am hoping to create a PR either today or tomorrow for the issue I am working on.
    Ladislav Thon
    @Ladicek
    I don't have anything to discuss either
    Carlos Andres De La Rosa
    @carlosdlr
    me either just the PR and i saw your comments just work on that
    Ladislav Thon
    @Ladicek
    I was busy last few days hunting down a concurrency problem in SmallRye Fault Tolerance, but I've managed to finish that today, so hopefully this week I'll be able to get to the issue I have assigned (circuit breaker / bulkhead lifecycle)
    actually if we cancel the call today, I might as well use the hour to start working on it :-)
    Emily Jiang
    @Emily-Jiang
    cancelled today and hope to see more progress before our next call next week including mine :o
    Andrew Rouse
    @Azquelt
    I wrote up a proposal for the metrics tags, as well as some open questions in #401
    Emily Jiang
    @Emily-Jiang
    Since there are a few tck changes made recently, I am keen to produce a macro release FT 2.0.1. @jbee have you got all TCK issues sorted?
    Jan Bernitt
    @jbee
    @Emily-Jiang Yes, thanks for checking.
    Andrew Rouse
    @Azquelt
    I've created a 2.1.x branch so we can start backporting any fixes for a 2.1.1 release.
    Emily Jiang
    @Emily-Jiang
    ok thanks @jbee @Azquelt
    post a message here once all PRs are ready so that I can kick off a build