Channels
  • s

    seph

    4 months ago
    @Stefano Bonicatti I think I’m ready to cut 5.2.2. Do you want to try to come to resolution on #7474 before?
  • Stefano Bonicatti

    Stefano Bonicatti

    4 months ago
    There doesn’t seem to be consensus about that choice and we really have spoken briefly me and you about that, so I guess that kind of choice can be done on a blueprint. I’m still slightly concerned though by the high default graceful time period, which is really meant for the normal shutdown workflow. Would a solution where we keep the “force” logic for the resource limit kill, but we wait 4 seconds for the graceful kill and then forcefully kill. 4 seconds is about the time osquery was waiting before we have changed the alarm timeout.
    This way osquery has “some” time to shutdown cleanly, but we also don’t wait too much. And it’s lower than what it’s currently achievable with lowering the
    alarm_timeout
    , which should be left alone.
  • s

    seph

    4 months ago
    Maybe a 4s grace is enough? That feels like enough that an unwedged process should be able to checkpoint.
    I feel like it would be to get some fix in, but also that it’s a bug that predates 5.2, so maybe we don’t need to hold a prerelease for it.
  • Stefano Bonicatti

    Stefano Bonicatti

    4 months ago
    Well, lets shift all of this to the 5.3.0 then
  • s

    seph

    4 months ago
    Okay! I’ll cut 5.2.2 later today, and we’ll see if we like it