gluster-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Gluster-devel] add-brick


From: Anand Avati
Subject: Re: [Gluster-devel] add-brick
Date: Mon, 20 Aug 2012 01:47:40 -0700

Something like this: (not tested at all)

diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c
index 4d8d326..0f8322f 100644
--- a/xlators/cluster/dht/src/dht-common.c
+++ b/xlators/cluster/dht/src/dht-common.c
@@ -191,7 +191,7 @@ dht_discover_complete (xlator_t *this, call_frame_t *discover_frame)
                                 "ENOENT errors: %d)", local->loc.path,
                                 (ret < 0) ? "yes" : "no", (ret > 0) ? ret : 0);
                         op_errno = EINVAL;
-                        goto out;
+                        goto selfheal;
                 }
 
                 dht_layout_set (this, local->inode, layout);
@@ -206,6 +206,12 @@ out:
                           NULL);
 
         return ret;
+selfheal:
+        FRAME_SU_DO (frame, dht_local_t);
+        uuid_copy (local->loc.gfid, local->gfid);
+        ret = dht_selfheal_directory (frame, dht_lookup_selfheal_cbk,
+                                      &local->loc, layout);
+       return 0;
 }


Inside dht_selfheal_directory and beyond we would need changes to make it work specially for nameless loc_t where it would not try to create missing dirs but only fix layout.

All this is necessary only for fixing 1-brick -> 2-brick kind of add-brick (which should be broken even on Linux now)

Avati

On Mon, Aug 20, 2012 at 12:38 AM, Emmanuel Dreyfus <address@hidden> wrote:
On Sun, Aug 19, 2012 at 06:38:48AM -0700, Anand Avati wrote:
> I now realize that if you re-refer an inode before it is timed out in the
> kernel (and LOOKUP is sent), then the code path
> hitting fuse_resolve_gfid_cbk might reach this. We need to call
> dht_self_heal in dht_discover_cbk if layout requires healing.

There are many flavors of dht_selfheal_*(), the change is not
straightforward.  Would you have a patch (even untested, just to
give the direction)?

--
Emmanuel Dreyfus
address@hidden


reply via email to

[Prev in Thread] Current Thread [Next in Thread]