[geo-rep]: Worker still ACTIVE after killing bricks

Problem: In changelog xlator after destroying listener it call's
         unlink to delete changelog socket file but socket file
         reference is not cleaned up from process memory

Solution: 1) To cleanup reference completely from process memory
             serialize transport cleanup for changelog and then
             unlink socket file
          2) Brick xlator will notify GF_EVENT_PARENT_DOWN to next
             xlator only after cleanup all xprts

Test: To test the same run below steps
      1) Setup some volume and enable brick mux
      2) kill anyone brick with gf_attach
      3) check changelog socket for specific to killed brick
         in lsof, it should cleanup completely

fixes: bz#1600145

Change-Id: Iba06cbf77d8a87b34a60fce50f6d8c0d427fa491
Signed-off-by: Mohit Agrawal <moagrawal@redhat.com>

1 files changed, 38 insertions, 1 deletions

diff --git a/xlators/features/changelog/src/changelog.c b/xlators/features/changelog/src/changelog.c
index 1f42dbe178e..1f22a97a6e9 100644
--- a/xlators/features/changelog/src/changelog.c
+++ b/xlators/features/changelog/src/changelog.c
@@ -2004,6 +2004,10 @@ notify(xlator_t *this, int event, void *data, ...)
     struct list_head queue = {
         0,
     };
+    uint64_t xprtcnt = 0;
+    uint64_t clntcnt = 0;
+    changelog_clnt_t *conn = NULL;
+    gf_boolean_t cleanup_notify = _gf_false;
 
     INIT_LIST_HEAD(&queue);
 
@@ -2011,6 +2015,33 @@ notify(xlator_t *this, int event, void *data, ...)
     if (!priv)
         goto out;
 
+    if (event == GF_EVENT_PARENT_DOWN) {
+        priv->victim = data;
+        gf_log(this->name, GF_LOG_INFO,
+               "cleanup changelog rpc connection of brick %s",
+               priv->victim->name);
+
+        this->cleanup_starting = 1;
+        changelog_destroy_rpc_listner(this, priv);
+        conn = &priv->connections;
+        if (conn)
+            changelog_ev_cleanup_connections(this, conn);
+        xprtcnt = GF_ATOMIC_GET(priv->xprtcnt);
+        clntcnt = GF_ATOMIC_GET(priv->clntcnt);
+
+        if (!xprtcnt && !clntcnt) {
+            LOCK(&priv->lock);
+            {
+                cleanup_notify = priv->notify_down;
+                priv->notify_down = _gf_true;
+            }
+            UNLOCK(&priv->lock);
+            if (!cleanup_notify)
+                default_notify(this, GF_EVENT_PARENT_DOWN, data);
+        }
+        goto out;
+    }
+
     if (event == GF_EVENT_TRANSLATOR_OP) {
         dict = data;
 
@@ -2629,8 +2660,10 @@ static void
 changelog_cleanup_rpc(xlator_t *this, changelog_priv_t *priv)
 {
     /* terminate rpc server */
-    changelog_destroy_rpc_listner(this, priv);
+    if (!this->cleanup_starting)
+        changelog_destroy_rpc_listner(this, priv);
 
+    (void)changelog_cleanup_rpc_threads(this, priv);
     /* cleanup rot buffs */
     rbuf_dtor(priv->rbuf);
 
@@ -2703,6 +2736,10 @@ init(xlator_t *this)
 
     LOCK_INIT(&priv->lock);
     LOCK_INIT(&priv->c_snap_lock);
+    GF_ATOMIC_INIT(priv->listnercnt, 0);
+    GF_ATOMIC_INIT(priv->clntcnt, 0);
+    GF_ATOMIC_INIT(priv->xprtcnt, 0);
+    INIT_LIST_HEAD(&priv->xprt_list);
 
     ret = changelog_init_options(this, priv);
     if (ret)