Copied to clipboard

Flag this post as spam?

This post will be reported to the moderators as potential spam to be looked at


  • David Veksler 81 posts 166 karma points
    May 17, 2016 @ 21:30
    David Veksler
    0

    Load-balanced Umbraco restarts when publishing content under high load

    When traffic is high, (re)publishing any page often causes all 3 applications pools (master + 2 read nodes) to restart.

    This causes the load balancer (which monitors ping.aspx) to mark all our nodes as offline, which results in a very bad user experience. We are having to tell editors not to use the Back Office during high traffic times. We can't reproduce this on our (single) dev environment. I can see that App_Data\umbraco.config goes to 0 then the normal size - maybe this is by design.

    I don't know how to debug this issue. We've had several experienced Umbraco developers look at this, and they don't know either.

    We're using Umbraco 7.4.3. Architecture diagram below.

    I think this event corresponds to most of the outages:

    2016-05-16 01:48:07,638 [P9432/D2/T1] INFO  Umbraco.Core.DatabaseContext - CanConnect = True
     2016-05-16 01:48:30,497 [P7384/D2/T4] INFO  Umbraco.Core.UmbracoApplicationBase - Application shutdown. Details: HostingEnvironment
    
    _shutDownMessage=HostingEnvironment initiated shutdown
    HostingEnvironment caused shutdown
    
    _shutDownStack=   at System.Environment.GetStackTrace(Exception e, Boolean needFileInfo)
       at System.Environment.get_StackTrace()
       at System.Web.Hosting.HostingEnvironment.InitiateShutdownInternal()
       at System.Web.Hosting.HostingEnvironment.InitiateShutdownWithoutDemand()
       at System.Web.Hosting.PipelineRuntime.StopProcessing()
    

    enter image description here

  • Jeavon Leopold 3074 posts 13632 karma points MVP 11x admin c-trib
    May 17, 2016 @ 21:36
    Jeavon Leopold
    0

    Are your servers running 2012R2 and how big is your Umbraco.Config?

  • David Veksler 81 posts 166 karma points
    May 17, 2016 @ 21:52
    David Veksler
    0

    Yes, 2012 R2 and umbraco.config is 181MB

    I confirmed that this fix which fixes App Pool restarts is installed, along with all optional Windows updates.

  • Jeavon Leopold 3074 posts 13632 karma points MVP 11x admin c-trib
    May 17, 2016 @ 22:40
    Jeavon Leopold
    0

    Double check that you have KB3052480 installed. This is the exact behaviour that I have seen with unpatched 2102r2 servers with a Umbraco.config that is large enough to trigger the bug as it lives in App_Data

  • Nicholas Westby 2054 posts 7103 karma points c-trib
    May 17, 2016 @ 22:10
    Nicholas Westby
    0

    Is IIS configured to recycle the application pool when memory exceeds some specified limit?

    Are there any errors in the Umbraco log prior to the "HostingEnvironment initiated shutdown" message?

  • Nicholas Westby 2054 posts 7103 karma points c-trib
    May 17, 2016 @ 22:12
    Nicholas Westby
    0

    Also, you may want to try disabling rapid fail protection for the application pool in IIS:

    Rapid Fail Protection

  • David Veksler 81 posts 166 karma points
    Aug 03, 2016 @ 03:18
    David Veksler
    0

    Hi guys,

    I have configured the above App Pool settings and triple-checked the patch install, but the issue is still happening.

    When new nodes are published under load, the umbraco.config will go to 0, the application pool will stop responding on all three nodes. The website will become unavailable because the load balancer all three servers go offline.

    I have checked the logs on the master and read-only servers, but I do not see error. I just see:

    2016-08-02 23:03:12,845 [P3004/D5/T63] INFO umbraco.content - Save Xml to file...

    2016-08-02 23:04:59,691 [P3004/D5/T60] INFO umbraco.content - Saved Xml to file.

    Any ideas?

  • David Veksler 81 posts 166 karma points
    Aug 03, 2016 @ 03:26
    David Veksler
    0

    By the way, when this happens, the Umbraco Back Office on the master node remains fine (since it does not use the cache). The read-only Back Offices do go down (the application pool is busy).

    I checked the Windows Event Log and it does not seem that the application pool is recycling - it's just busy rebuilding the 200MB cache each time.

Please Sign in or register to post replies

Write your reply to:

Draft