then we may not saywe support suspend on the front page ;-) at least, it misleading me
On 2015-04-21 00:43 , Sumit Mohanty Wrote: There is no suspend if you are thinking in terms of VM pause/suspend. In case of Slider its a process kill. Slider preserves all the configuration and uses them to start when you start the application after stop. Stop is a kill on the application processes. If application has an implementation of the stop() command it may get called but there is no guarantee that stop() command will be allowed to complete before app process is killed. On Mon, Apr 20, 2015 at 6:51 AM, <[email protected]> wrote: > what.is slider suspend app logic? > > > > On 2015-04-20 17:41 , Steve Loughran Wrote: > > > > On 19 Apr 2015, at 03:57, [email protected] wrote: > > > > does slider support suspend application? > > sigstop? > > > > > > container release without warning. > > > > > > > > On 2015-04-18 21:33 , Steve Loughran Wrote: > > > > hi > > > > you can stop and application and start it again; when you stop the > application it kills the HBase processes and the slider AM, returning all > resources to YARN. > > > > If you really need to stop HBase for a while, that's the way to do it. > > > > you can also shrink a running application via the slider fiex command: > you can, say, shrink the number of HBase region servers. Each of these is > going to take on more work -so more CPU and network traffic; nothing comes > for free. But if your HBase instance isn't too busy, that works. > > > > Finally, you can use YARN pre-emption and tell slider to use a > pre-emptible queue for containers (you can do this in its resources.json > file), or just launch the application against the specific queue and have > it default to everywhere. I'd recommend having the AM non-preemptible, and > the other components pre-emptible. > > > > When that's set up, when higher priority work comes in, the HBase > containers would be killed. Slider will react by incrementing some failure > counters and asking for it back; it will run a smaller cluster until the > containers are allocated. If you try that, because Slider doesn't know that > a container was pre-empted (only that it exited), you'll need to edit some > of the failure threshold options in resources.json, stopping slider > over-reacting to these failures and killing the application > > > > -steve > > > > > >> On 18 Apr 2015, at 12:35, skaterQiang <[email protected]> wrote: > >> > >> Hello guys: > >> In the http://slider.incubator.apache.org/index.html it said > slider can > >> " > >> > >> Stop / Suspend / Resume application instances as needed > >> > >> " > >> and I really need this feature for suspend some job in my yarn cluster > and let some urgent jobs to have more resource. > >> But then I download the slider, it can only stop start "hbase, > memcached, storm and accumulo", and from the code, it use start and stop > command for hbase. > >> I think it is provisioning and cluster management not application > management. > >> > >> > >> Could slider suspend and resume yarn applications, or slider can > suspend and resume hbase or storm? > >> > >> > >> Thanks in advance, > >> Skater > > > > -- thanks Sumit
