cancel does not work on jobs in "runrequest" state

[trws]

Normally I don't think this would be a big deal, but somehow jobs every once in a while get into runrequest then never make progress again. Having a way to clear out such jobs from sched and wreck would be useful for such cases.

[dongahn]

`-f` doesn't kill it?

…

________________________________ From: Tom Scogland <[email protected]> Sent: Sunday, April 8, 2018 9:24:12 PM To: flux-framework/flux-core Cc: Subscribed Subject: [flux-framework/flux-core] cancel does not work on jobs in "runrequest" state (#1438) Normally I don't think this would be a big deal, but somehow jobs every once in a while get into runrequest then never make progress again. Having a way to clear out such jobs from sched and wreck would be useful for such cases. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<#1438>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AA0nq3HMHe-w27Fwl4mBd27XqeKKHnjlks5tmuJsgaJpZM4TL62C>.

[trws]

I’ll try that next time it comes up, but I don’t think so.

…

On 9 Apr 2018, at 8:15, Dong H. Ahn wrote: `-f` doesn't kill it? ________________________________ From: Tom Scogland ***@***.***> Sent: Sunday, April 8, 2018 9:24:12 PM To: flux-framework/flux-core Cc: Subscribed Subject: [flux-framework/flux-core] cancel does not work on jobs in "runrequest" state (#1438) Normally I don't think this would be a big deal, but somehow jobs every once in a while get into runrequest then never make progress again. Having a way to clear out such jobs from sched and wreck would be useful for such cases. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<#1438>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AA0nq3HMHe-w27Fwl4mBd27XqeKKHnjlks5tmuJsgaJpZM4TL62C>. -- You are receiving this because you authored the thread. Reply to this email directly or view it on GitHub: #1438 (comment)

[grondo]

In the current system "canceling" a job in the runrequest state is going to be difficult if not impossible. It is a possible flaw in the engineering of the way that jobs are launched, and we should design the "starting" process better next time so this is possible. One problem with the current design is that in `runrequest` state until we reach `running` I'm not sure that wrexecd daemons are listening on their flux handles so they may not even hear that a starting job should be canceled (likely they are blocked on a barrier, so it is conceivable to fix this in the job shell replacement). On Mon, Apr 9, 2018 at 8:20 AM, Tom Scogland <[email protected]> wrote:

…

I’ll try that next time it comes up, but I don’t think so. On 9 Apr 2018, at 8:15, Dong H. Ahn wrote: > `-f` doesn't kill it? > ________________________________ > From: Tom Scogland ***@***.***> > Sent: Sunday, April 8, 2018 9:24:12 PM > To: flux-framework/flux-core > Cc: Subscribed > Subject: [flux-framework/flux-core] cancel does not work on jobs in > "runrequest" state (#1438) > > > Normally I don't think this would be a big deal, but somehow jobs > every once in a while get into runrequest then never make progress > again. Having a way to clear out such jobs from sched and wreck would > be useful for such cases. > > — > You are receiving this because you are subscribed to this thread. > Reply to this email directly, view it on > GitHub<#1438>, or > mute the > thread<https://github.com/notifications/unsubscribe-auth/AA0nq3HMHe- w27Fwl4mBd27XqeKKHnjlks5tmuJsgaJpZM4TL62C>. > > > -- > You are receiving this because you authored the thread. > Reply to this email directly or view it on GitHub: > https://github.com/flux-framework/flux-core/issues/ 1438#issuecomment-379788208 — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1438 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAtSUuOWAEiFewU16VHzHtB_r3R5Rm1Uks5tm3xKgaJpZM4TL62C> .

[grondo]

closed by #1988