shanthoosh commented on issue #1241: SAMZA-2424: AM should cache and serve serialized Job Model to containers URL: https://github.com/apache/samza/pull/1241#issuecomment-568138070 Please correct me if i'm wrong. JobModel currently contains the container-to-host-locality information, accessible via the high-level field `all-container-locality`. In a typical run of a samza-job, containers could move around different NM hosts due to transient h/w failures (or containers could be explicitly moved around by using our powerful container placement API) So if we cache the JobModel and after that the container-locality changes, then subsequent AM HTTP reads of JobModel could potentially be serving inconsistent JobModel. Sample JobModel taken from a samza job ```bash [svenkata@host ~]$ cat ~/job-model | jq . | head -n 10 { "all-container-locality": { "68-0": "host-1", "135-0": "host-2", "173-0": "host-3", "158-0": "host-4", "45-0": "host-5", "112-0": "host-6", "109": "host-7", "108": "host-8", ``` In the above sample JobModel, the active container '68-0' runs in host-1. Currently the ApplicationMaster process is used as the source of truth for JobModel. Some of our internal samza-admin monitors uses it to garbage-collect the left-over local RocksDB state-stores from disk(to reclaim disk-space). Just curious, if the above API change would affect our state-store GC-monitor.
---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
