Re: [Openstack-operators] [nova] nova-compute automatically disabling itself?
On 2/6/2018 6:44 PM, Matt Riedemann wrote: On 2/6/2018 2:14 PM, Chris Apsey wrote: but we would rather have intermittent build failures rather than compute nodes falling over in the future. Note that once a compute has a successful build, the consecutive build failures counter is reset. So if your limit is the default (10) and you have 10 failures in a row, the compute service is auto-disabled. But if you have say 5 failures and then a pass, it's reset to 0 failures. Obviously if you're doing a pack-first scheduling strategy rather than spreading instances across the deployment, a burst of failures could easily disable a compute, especially if that host is overloaded like you saw. I'm not sure if rescheduling is helping you or not - that would be useful information since we consider the need to reschedule off a failed compute host as a bad thing. At the Forum in Boston when this idea came up, it was specifically for the case that operators in the room didn't want a bad compute to become a "black hole" in their deployment causing lots of reschedules until they get that one fixed. Just an update on this. There is a change merged in Rocky [1] which is also going through backports to Queens and Pike. If you've already disabled the "consecutive_build_service_disable_threshold" config option then it's a no-op. If you haven't, "consecutive_build_service_disable_threshold" is now used to count build failures but no longer auto-disable the compute service on the configured threshold is met (10 by default). The build failure count is then used by a new weigher (enabled by default) to sort hosts with build failures to the back of the list of candidate hosts for new builds. Once there is a successful build on a given host, the failure count is reset. The idea here is that hosts which are failing are given lower priority during scheduling. [1] https://review.openstack.org/#/c/572195/ -- Thanks, Matt ___ OpenStack-operators mailing list OpenStack-operators@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
Re: [Openstack-operators] [nova] nova-compute automatically disabling itself?
On 2/6/2018 2:14 PM, Chris Apsey wrote: but we would rather have intermittent build failures rather than compute nodes falling over in the future. Note that once a compute has a successful build, the consecutive build failures counter is reset. So if your limit is the default (10) and you have 10 failures in a row, the compute service is auto-disabled. But if you have say 5 failures and then a pass, it's reset to 0 failures. Obviously if you're doing a pack-first scheduling strategy rather than spreading instances across the deployment, a burst of failures could easily disable a compute, especially if that host is overloaded like you saw. I'm not sure if rescheduling is helping you or not - that would be useful information since we consider the need to reschedule off a failed compute host as a bad thing. At the Forum in Boston when this idea came up, it was specifically for the case that operators in the room didn't want a bad compute to become a "black hole" in their deployment causing lots of reschedules until they get that one fixed. -- Thanks, Matt ___ OpenStack-operators mailing list OpenStack-operators@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
Re: [Openstack-operators] [nova] nova-compute automatically disabling itself?
All, This was the core issue - setting consecutive_build_service_disable_threshold = 0 in nova.conf (on controllers and compute nodes) solved this. It was being triggered by neutron dropping requests (and/or responses) for vif-plugging due to cpu usage on the neutron endpoints being pegged at 100% for too long. We increased our rpc_response_timeout value and this issue appears to be resolved for the time being. We can probably safely remove the consecutive_build_service_disable_threshold option at this point, but we would rather have intermittent build failures rather than compute nodes falling over in the future. Slightly related, we are noticing that neutron endpoints are using noticeably more CPU time recently than in the past w/ a similar workload (we run linuxbridge w/ vxlan). We believe this is tied to our application of KPTI for meltdown mitigation across the various hosts in our cluster (the timeline matches). Has anyone else experienced similar impacts or can suggest anything to try to lessen the impact? --- v/r Chris Apsey bitskr...@bitskrieg.net https://www.bitskrieg.net On 2018-01-31 04:47 PM, Chris Apsey wrote: That looks promising. I'll report back to confirm the solution. Thanks! --- v/r Chris Apsey bitskr...@bitskrieg.net https://www.bitskrieg.net On 2018-01-31 04:40 PM, Matt Riedemann wrote: On 1/31/2018 3:16 PM, Chris Apsey wrote: All, Running in to a strange issue I haven't seen before. Randomly, the nova-compute services on compute nodes are disabling themselves (as if someone ran openstack compute service set --disable hostX nova-compute. When this happens, the node continues to report itself as 'up' - the service is just disabled. As a result, if enough of these occur, we get scheduling errors due to lack of available resources (which makes sense). Re-enabling them works just fine and they continue on as if nothing happened. I looked through the logs and I can find the API calls where we re-enable the services (PUT /v2.1/os-services/enable), but I do not see any API calls where the services are getting disabled initially. Is anyone aware of any cases where compute nodes will automatically disable their nova-compute service on their own, or has anyone seen this before and might know a root cause? We have plenty of spare vcpus and RAM on each node - like less than 25% utilization (both in absolute terms and in terms of applied ratios). We're seeing follow-on errors regarding rmq messages getting lost and vif-plug failures, but we think those are a symptom, not a cause. Currently running pike on Xenial. --- v/r Chris Apsey bitskr...@bitskrieg.net https://www.bitskrieg.net ___ OpenStack-operators mailing list OpenStack-operators@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators This is actually a feature added in Pike: https://review.openstack.org/#/c/463597/ This came up in discussion with operators at the Forum in Boston. The vif-plug failures are likely the reason those computes are getting disabled. There is a config option "consecutive_build_service_disable_threshold" which you can set to disable the auto-disable behavior as some have experienced issues with it: https://bugs.launchpad.net/nova/+bug/1742102 ___ OpenStack-operators mailing list OpenStack-operators@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators ___ OpenStack-operators mailing list OpenStack-operators@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
Re: [Openstack-operators] [nova] nova-compute automatically disabling itself?
That looks promising. I'll report back to confirm the solution. Thanks! --- v/r Chris Apsey bitskr...@bitskrieg.net https://www.bitskrieg.net On 2018-01-31 04:40 PM, Matt Riedemann wrote: On 1/31/2018 3:16 PM, Chris Apsey wrote: All, Running in to a strange issue I haven't seen before. Randomly, the nova-compute services on compute nodes are disabling themselves (as if someone ran openstack compute service set --disable hostX nova-compute. When this happens, the node continues to report itself as 'up' - the service is just disabled. As a result, if enough of these occur, we get scheduling errors due to lack of available resources (which makes sense). Re-enabling them works just fine and they continue on as if nothing happened. I looked through the logs and I can find the API calls where we re-enable the services (PUT /v2.1/os-services/enable), but I do not see any API calls where the services are getting disabled initially. Is anyone aware of any cases where compute nodes will automatically disable their nova-compute service on their own, or has anyone seen this before and might know a root cause? We have plenty of spare vcpus and RAM on each node - like less than 25% utilization (both in absolute terms and in terms of applied ratios). We're seeing follow-on errors regarding rmq messages getting lost and vif-plug failures, but we think those are a symptom, not a cause. Currently running pike on Xenial. --- v/r Chris Apsey bitskr...@bitskrieg.net https://www.bitskrieg.net ___ OpenStack-operators mailing list OpenStack-operators@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators This is actually a feature added in Pike: https://review.openstack.org/#/c/463597/ This came up in discussion with operators at the Forum in Boston. The vif-plug failures are likely the reason those computes are getting disabled. There is a config option "consecutive_build_service_disable_threshold" which you can set to disable the auto-disable behavior as some have experienced issues with it: https://bugs.launchpad.net/nova/+bug/1742102 ___ OpenStack-operators mailing list OpenStack-operators@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
Re: [Openstack-operators] [nova] nova-compute automatically disabling itself?
There's [1], but I would have expected you to see error logs like [2] if that's what you're hitting. [1] https://github.com/openstack/nova/blob/master/nova/conf/compute.py#L627-L645 [2] https://github.com/openstack/nova/blob/master/nova/compute/manager.py#L1714-L1716 efried On 01/31/2018 03:16 PM, Chris Apsey wrote: > All, > > Running in to a strange issue I haven't seen before. > > Randomly, the nova-compute services on compute nodes are disabling > themselves (as if someone ran openstack compute service set --disable > hostX nova-compute. When this happens, the node continues to report > itself as 'up' - the service is just disabled. As a result, if enough > of these occur, we get scheduling errors due to lack of available > resources (which makes sense). Re-enabling them works just fine and > they continue on as if nothing happened. I looked through the logs and > I can find the API calls where we re-enable the services (PUT > /v2.1/os-services/enable), but I do not see any API calls where the > services are getting disabled initially. > > Is anyone aware of any cases where compute nodes will automatically > disable their nova-compute service on their own, or has anyone seen this > before and might know a root cause? We have plenty of spare vcpus and > RAM on each node - like less than 25% utilization (both in absolute > terms and in terms of applied ratios). > > We're seeing follow-on errors regarding rmq messages getting lost and > vif-plug failures, but we think those are a symptom, not a cause. > > Currently running pike on Xenial. > > --- > v/r > > Chris Apsey > bitskr...@bitskrieg.net > https://www.bitskrieg.net > > ___ > OpenStack-operators mailing list > OpenStack-operators@lists.openstack.org > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators ___ OpenStack-operators mailing list OpenStack-operators@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
Re: [Openstack-operators] [nova] nova-compute automatically disabling itself?
On 1/31/2018 3:16 PM, Chris Apsey wrote: All, Running in to a strange issue I haven't seen before. Randomly, the nova-compute services on compute nodes are disabling themselves (as if someone ran openstack compute service set --disable hostX nova-compute. When this happens, the node continues to report itself as 'up' - the service is just disabled. As a result, if enough of these occur, we get scheduling errors due to lack of available resources (which makes sense). Re-enabling them works just fine and they continue on as if nothing happened. I looked through the logs and I can find the API calls where we re-enable the services (PUT /v2.1/os-services/enable), but I do not see any API calls where the services are getting disabled initially. Is anyone aware of any cases where compute nodes will automatically disable their nova-compute service on their own, or has anyone seen this before and might know a root cause? We have plenty of spare vcpus and RAM on each node - like less than 25% utilization (both in absolute terms and in terms of applied ratios). We're seeing follow-on errors regarding rmq messages getting lost and vif-plug failures, but we think those are a symptom, not a cause. Currently running pike on Xenial. --- v/r Chris Apsey bitskr...@bitskrieg.net https://www.bitskrieg.net ___ OpenStack-operators mailing list OpenStack-operators@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators This is actually a feature added in Pike: https://review.openstack.org/#/c/463597/ This came up in discussion with operators at the Forum in Boston. The vif-plug failures are likely the reason those computes are getting disabled. There is a config option "consecutive_build_service_disable_threshold" which you can set to disable the auto-disable behavior as some have experienced issues with it: https://bugs.launchpad.net/nova/+bug/1742102 -- Thanks, Matt ___ OpenStack-operators mailing list OpenStack-operators@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators