Automic Workload Automation

 View Only
  • 1.  Independent Ram memory allocation for WP in automic CDA Continuous Delivery Automation. 12.2

    Posted Mar 19, 2020 05:59 PM

    Good morning, Community,


    It is possible to independently allocate RAM to each of the WPs (work processes) to increase processing without having many WPs running in parallel in the service manager

    Thank you very much.



  • 2.  RE: Independent Ram memory allocation for WP in automic CDA Continuous Delivery Automation. 12.2

    Posted Mar 23, 2020 10:15 AM
    Hi.

    As far ​as I know, no, not with Automic.

    You could certainly use Linux cgroups or some such facility to limit RAM on a per-process basis, but this might not be a great idea.

    Best,
    Carsten


  • 3.  RE: Independent Ram memory allocation for WP in automic CDA Continuous Delivery Automation. 12.2

    Posted Mar 24, 2020 11:37 AM
    Hi,

    How would I know how many WP's my CDA needs?


  • 4.  RE: Independent Ram memory allocation for WP in automic CDA Continuous Delivery Automation. 12.2
    Best Answer

    Posted Mar 24, 2020 11:47 AM
    Hi,

    I don't know of any surefire way. It depends on a number of things - number of executions, types of jobs, probably even the context switching capabilities of the underlying OS and also the Automic version.

    We currently run 20 WP (four of which DWP and one PWP, naturally), plus four JWP and about ten CP processes on our Linux production system with V12.x.

    We arrived at this number somewhat by trial and error. At some point we ran a double amount of WP because Automic Support said so, but that turned out to be too many. When you increase the number of WP too high, coupled with the "right" types of jobs or scripts, in my experience, you increase the risk for database deadlocks and might see more of the accompanying log messages about long-running DB queries, and generally diminished performance. Ultimately, it's a bit of trial and error, helped by looking at log files (or sizes) for the processes to judge how much work they are pulling, and and the load graphs in Automic.

    Hth,