03652e7497
Prior to this change, if the persistent tasks framework noticed that a job was running on a node that was isolated but has rejoined the cluster then it would close that job. This was not ideal, because then the job would persist state from the autodetect process that was isolated. This commit changes the behaviour to kill the autodetect process associated with such a job, so that it does not interfere with the autodetect process that is running on the node where the persistent tasks framework thinks it should be running. In order to achieve this a change has also been made to the behaviour of force-close. Previously this would result in the autodetect process being gracefully shut down asynchronously to the force-close request. However, the mechanism by which this happened was the same as the mechanism for cancelling tasks that end up running on more than one node due to nodes becoming isolated from the cluster. Therefore, force-close now also kills the autodetect process rather than gracefully stopping it. The documentation has been changed to reflect this. It should not be a problem as force-close is supposed to be a last resort for when normal close fails. relates elastic/x-pack-elasticsearch#1186 Original commit: elastic/x-pack-elasticsearch@578c944371 |
||
|---|---|---|
| .. | ||
| graph | ||
| ml | ||
| security | ||
| watcher | ||
| defs.asciidoc | ||
| graph.asciidoc | ||
| index.asciidoc | ||
| ml-api.asciidoc | ||
| security.asciidoc | ||
| watcher.asciidoc | ||
| xindex.asciidoc | ||