Bug 1564918
Summary: | On ipmitool failure, it doesn't look like ironic is retrying eventhough it says after "4 retry" | ||
---|---|---|---|
Product: | Red Hat OpenStack | Reporter: | David Hill <dhill> |
Component: | openstack-ironic | Assignee: | Ilya Etingof <ietingof> |
Status: | CLOSED ERRATA | QA Contact: | bjacot |
Severity: | medium | Docs Contact: | |
Priority: | medium | ||
Version: | 10.0 (Newton) | CC: | bfournie, ietingof, jkreger, joflynn, jschluet, mburns, rhel-osp-director-maint, srevivo |
Target Milestone: | z2 | Keywords: | Triaged, ZStream |
Target Release: | 13.0 (Queens) | ||
Hardware: | Unspecified | ||
OS: | Unspecified | ||
Whiteboard: | |||
Fixed In Version: | openstack-ironic-10.1.3-2.el7ost | Doc Type: | Enhancement |
Doc Text: |
Previously, Ironic considered just one IPMI error as retryable. That might have caused unjustified Ironic failure. With this enhancement, Ironic treats more types of IPMI error messages as retryable by the IPMI-backed hardware interfaces, such as power and management hardware interfaces. Specifically, "Node busy", "Timeout", "Out of space", and "BMC initialization in progress" IPMI errors cause Ironic to retry the IPMI command. The result is improved reliability of IPMI based communication with BMC.
|
Story Points: | --- |
Clone Of: | Environment: | ||
Last Closed: | 2018-08-29 16:32:24 UTC | Type: | Bug |
Regression: | --- | Mount Type: | --- |
Documentation: | --- | CRM: | |
Verified Versions: | Category: | --- | |
oVirt Team: | --- | RHEL 7.3 requirements from Atomic Host: | |
Cloudforms Team: | --- | Target Upstream Version: | |
Embargoed: |
Description
David Hill
2018-04-08 22:53:52 UTC
Greetings, I think retry logic in vbmc is likely the right course of action to take. The reason being is that ipmitool only retries when there are cases of no reply received from the BMC. Vbmc is in essence indicating that there has been a hard failure powering on the "node", at which point ipmitool and ironic don't try any sort of retry operation. The best way to think about it is connectivity or transitory issue retry as opposed to hard failure retries. Hopefully that makes sense! -Julia It also seems that Ironic's notion of retryable failures is limited to a single `lanplus` specific error message. Thus, it may make sense to hardcode more error messages into Ironic to appropriate other error messages that `ipmitool` can potentially produce when reporting soft failures. Once Ironic retries on soft errors like "timeout" or "node busy", we could make VirtualBMC reporting one of these soft errors on libvirt faults what finally brings all the pieces together. I think it makes sense to retry on more error conditions because in the field, various BMCs may behave differently. For error messages see [1], particularly generic "completion codes" (p.44) and "device specific codes" (p.154). 1. https://www.intel.com/content/dam/www/public/us/en/documents/product-briefs/ipmi-second-gen-interface-spec-v2-rev1-1.pdf Is there any reasons for not simply retrying on every kind of failures ? I'd expect the retry to always retry unless it's in the specifications... and if a specification says "try only once if this error is returned" , I'd be saying it's a bad specification. > Is there any reasons for not simply retrying on every kind of failures?
I so not think they have specific instructions in the spec on the retries and timing. But they enumerate many errors and some of them look "softer" e.g. retryable.
Consider for example "authentication failure" versus "BMC initialization in progress" errors. To me it feels like there is no much point in pondering unsuccessful authentication (you can even get blacklisted at the BMC if you do), but it does make sense to wait a little for BMC to come up and try again.
Also, retrying implies wasted time. It may make sense to fail fast if we have little hope to recover.
Installed OSP13 cat /etc/yum.repos.d/latest-installed 13 -p 2018-07-30.2 verified that /ironic/drivers/modules/ipmitool.py had latest gerrit changes. Changes are included. Please reopen if you experience this problem again. This bug is marked for inclusion in the errata but does not currently contain draft documentation text. To ensure the timely release of this advisory please provide draft documentation text for this bug as soon as possible. If you do not think this bug requires errata documentation, set the requires_doc_text flag to "-". To add draft documentation text: * Select the documentation type from the "Doc Type" drop down field. * A template will be provided in the "Doc Text" field based on the "Doc Type" value selected. Enter draft text in the "Doc Text" field. May be this sentence "Ironic fails on potentially recoverable IPMI failures." should be re-phrased to indicate that this is what has been happening *prior* to this erratum. I think the logic is inverse here. Before this enhancement ironic considered just one IPMI error as retryable. That may have caused unjustified ironic failure. This enhancement made ironic considering more types of IPMI errors as recoverable e.g. ironic would retry the failed command if the command failed with one of those now retryable errors. Does it make sense? Thanks Ilya Etingof Thanks Ilya Etingof Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory, and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://access.redhat.com/errata/RHBA-2018:2593 |