wilfred-s commented on code in PR #560: URL: https://github.com/apache/yunikorn-site/pull/560#discussion_r3478741842
########## docs/user_guide/placement_rules.md: ########## @@ -285,6 +285,35 @@ Result: `root.testing` Application submit request for a non kubernetes based application by the user `developer`<br/> Result: failed, next rule executed +### Recovery Rule +Internal rule name: *recovery* + +The recovery rule is an internal rule that is automatically added as the _last_ rule of every placement policy. +It is implicit: it cannot be configured, reordered or removed and does not need to be added to the configuration. + +The rule only acts on applications that are submitted with the _force_ flag set. +The shim sets this flag for applications that own allocations which are already running on the cluster, for instance while the scheduler is being restarted or recovered. Review Comment: replace `own` with `contain` ########## docs/user_guide/placement_rules.md: ########## @@ -285,6 +285,35 @@ Result: `root.testing` Application submit request for a non kubernetes based application by the user `developer`<br/> Result: failed, next rule executed +### Recovery Rule +Internal rule name: *recovery* + +The recovery rule is an internal rule that is automatically added as the _last_ rule of every placement policy. +It is implicit: it cannot be configured, reordered or removed and does not need to be added to the configuration. Review Comment: `does not need to` should be `cannot` ########## docs/user_guide/placement_rules.md: ########## @@ -285,6 +285,35 @@ Result: `root.testing` Application submit request for a non kubernetes based application by the user `developer`<br/> Result: failed, next rule executed +### Recovery Rule +Internal rule name: *recovery* + +The recovery rule is an internal rule that is automatically added as the _last_ rule of every placement policy. +It is implicit: it cannot be configured, reordered or removed and does not need to be added to the configuration. + +The rule only acts on applications that are submitted with the _force_ flag set. +The shim sets this flag for applications that own allocations which are already running on the cluster, for instance while the scheduler is being restarted or recovered. +When such an application is not placed by any of the configured rules the recovery rule places it into the recovery queue `root.@recovery@`, so that the already running allocation can be recovered without user intervention. +Applications that do not have the force flag set are never placed in the recovery queue: for those applications the recovery rule returns no queue. + +The recovery queue is a dynamic, internally managed queue. +It has no quota and no ACL, it cannot be referenced by other rules and it cannot be submitted to directly. +See the [recovery queue](queue_config#recovery-queue) description for more details. + +Supported parameters: +* none + +## Handling placement failures +A rule only matches if it generates a queue that the application is allowed to run in. +A generated queue that does not exist and cannot be created, that is not a leaf queue, that is being drained, or that the user is not allowed to submit to is not a match. Review Comment: I would change this in the sentence `that is not a leaf queue,` to `already exists as a parent queue`. To explain more you can add a new sentence about the failure to create: create flag not set or a new leaf to be created under existing leaf -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
