zookeeper-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Pramod Srinivasan <pra...@juniper.net>
Subject Re: behavior of ephemeral nodes
Date Wed, 02 May 2018 19:21:12 GMT
Hi Andor,

Thanks for the response:

Below is the code snippet:

void
watcher(zhandle_t *zzh, int type, int state,
                     const char *path, void *ctx) {

    check_path(ctx);
}

void check_path(void *ctx) {
    struct String_vector children;
    int rc = 0;

   /* (1) get the current set of children for "/a" */
    rc = zoo_wget_children(zh, “/a”, watcher, ctx, &children);

    /* there are some children to “/a” */
    if (children.count > 0) {
        return;
    }

    /* (2) no children, try deleting “/a” */
    rc = zoo_delete(“/a”, -1);

    /* should not happen */
    if (rc == ZNOTEMPTY) {
        /* (3) get the current set of children for “/a” */
        rc = zoo_wget_children(zh, “/a”, watcher, ctx, &children);
  
        if (children.count == 0)
            assert(); /* we hit this assert */
    }
}

We have initiated watch of "/a" like below:

    rc = zoo_wget_children(zh, “/a”, watcher, ctx, &children);

Is it possible that the delete of ephemeral nodes are not committed yet by the Leader, hence
a delete (2) will fail. While a read returns that there are no nodes in (1) and (3) as the
session have expired and the znode delete has been initiated by Leader?

Thanks,
Pramod

On 5/2/18, 10:18 AM, "Andor Molnar" <andor@cloudera.com> wrote:

    Hi Pramod,
    
    I think you should be able to delete znode /a once both B and C sessions
    are closed/expired and childrens are removed.
    Would you please post relevant code snippets of your processes to get some
    insight on what's happening?
    
    Regards,
    Andor
    
    
    
    On Wed, May 2, 2018 at 7:11 PM Pramod Srinivasan <pramod@juniper.net> wrote:
    
    > any thoughts on this?
    >
    > On 5/1/18, 1:43 PM, "Pramod Srinivasan" <pramod@juniper.net> wrote:
    >
    >     Hello Everyone,
    >
    >     Can someone help me rationalize the behavior I see?
    >
    >     1) Process A is watching node /a
    >     2) Process B and Process C add ephemeral nodes, say /a/b and /a/c
    >     3) Process B and Process C close their session and Process A gets a
    > ZOO_CHILD_EVENT event when b and c goes away.
    >     4) In the context of ZOO_CHILD_EVENT, Process A does a
    > zoo_wget_children for “/a” and find its empty, but when it tries to delete
    > “/a”, it finds that it is not empty and zoo_wget_children again returns no
    > children.
    >
    >     How can get_children return no children for a znode and a remove of
    > the znode fail with not empty? It looks like there is a relation with
    > ephemeral nodes and Follower/Leader behavior, don’t understand this enough
    > to make sense
    >
    >     Thanks,
    >     Pramod
    >
    >
    >
    

Mime
View raw message