vSAN1

 View Only
  • 1.  How does vSAN EPD service work

    Posted Apr 30, 2022 05:46 AM

    I want to know what is vSAN EPD ?    I could just find this " EPD (Entry Persistence Daemon) is a user space daemon that runs on every host that is part of the vSAN cluster. The main job of EPD is to make sure there is no component leakage when objects are deleted.
    "

    Would you please give an example about this "there is no component leakage when objects are deleted."? What does mean components leakage ?



  • 2.  RE: How does vSAN EPD service work

    Broadcom Employee
    Posted Apr 30, 2022 05:48 AM

    Expect your thread to move to the vSAN area now I’ve reported it.

     



  • 3.  RE: How does vSAN EPD service work

    Posted Apr 30, 2022 06:23 AM

    Excuse me . It was my fault . How can I move it ? Do I have permission ?



  • 4.  RE: How does vSAN EPD service work

    Broadcom Employee
    Posted Apr 30, 2022 09:15 PM

    It has been moved now, check the path above your initial post.

     



  • 5.  RE: How does vSAN EPD service work
    Best Answer

    Posted May 01, 2022 10:20 PM

    , EPD is responsible for validation of 'garbage collection' e.g. that when components are marked for deletion that the blocks they were previously using are freed and that these data are properly dereferenced from all stored locations (in data and metadata).

     

    "component leakage" basically adheres to the same principals as memory leaks e.g. something in memory was supposed to be freed but was not (or not fully, e.g. dereferenced/cleared/not re-restored from somewhere else etc.). 

     

    A 'leaked' component would basically be something that has been marked for deletion/been deleted but is still stored and/or referenced somewhere (not necessarily in physical data-blocks but consider also metadata references, in caches etc.).

     

    This is a fairly backend daemon that rarely anyone would have any interest in unless they have a problem with it running - is there a functional reason you are asking about EPD?

    The most common issue related to EPD is the fact that it stores its local-DB in /scratch on nodes and if scratch is unavailable/full/late-to-arrive then it is not available to store this and EPD doesn't function - this isn't a major problem in the short term (hours-days (depending on workflow, specifically deletions)) but if left in this state for weeks-months then can become a problem for various reasons.