pnfs-obj: Proper delay for NFS4ERR_RECALLCONFLICT in layout_get_done

Commit Message

Boaz Harrosh Jan. 14, 2014, 2:58 p.m. UTC

Sorry forgot to CC Stable Tree <[e-mail]>

Greg hi

In Linux v3.9 there is a conflict around this area exactly do to change:
	[30005121] NFSv4.1: LAYOUTGET EDELAY loops timeout to the MDS

If there are stables below 3.9 please tell me I will send you a patch
for these.

Thanks
Boaz

-------- Original Message --------
Date: Tue, 14 Jan 2014 16:45:27 +0200
From: Boaz Harrosh <[e-mail]>
To: Trond Myklebust <[e-mail]>, NFS list <[e-mail]>, open-osd <[e-mail]>
Subject: [PATCH] pnfs-obj: Proper delay for NFS4ERR_RECALLCONFLICT in layout_get_done


An NFS4ERR_RECALLCONFLICT is returned by server from a GET_LAYOUT
only when a Server Sent a RECALL do to that GET_LAYOUT, or
the RECALL and GET_LAYOUT crossed on the wire.
In any way this means we want to wait at most until in-flight IO
is finished and the RECALL can be satisfied.

So a proper wait here is more like 1/10 of a second, not 15 seconds
like we have now. (We use NFS4_POLL_RETRY_MIN here)

Current code totally craps out performance of very large files on
most pnfs-objects layouts, because of how the map changes when the
file has grown and spills into the next raid group.

CC: Stable Tree <[e-mail]>
Signed-off-by: Boaz Harrosh <[e-mail]>
---
 fs/nfs/nfs4proc.c | 22 +++++++++++++++++++---
 1 file changed, 19 insertions(+), 3 deletions(-)

Comments

Boaz Harrosh Jan. 14, 2014, 3:10 p.m. UTC | #1

On 01/14/2014 04:58 PM, Boaz Harrosh wrote:
> Sorry forgot to CC Stable Tree <[e-mail]>
> 
> Greg hi
> 
> In Linux v3.9 there is a conflict around this area exactly do to change:
> 	[30005121] NFSv4.1: LAYOUTGET EDELAY loops timeout to the MDS
> 
> If there are stables below 3.9 please tell me I will send you a patch
> for these.
> 
> Thanks
> Boaz
> 
<>
> Subject: [PATCH] pnfs-obj: Proper delay for NFS4ERR_RECALLCONFLICT in layout_get_done
> 
> 
> An NFS4ERR_RECALLCONFLICT is returned by server from a GET_LAYOUT
> only when a Server Sent a RECALL do to that GET_LAYOUT, or
> the RECALL and GET_LAYOUT crossed on the wire.
> In any way this means we want to wait at most until in-flight IO
> is finished and the RECALL can be satisfied.
> 
> So a proper wait here is more like 1/10 of a second, not 15 seconds
> like we have now. (We use NFS4_POLL_RETRY_MIN here)
> 
> Current code totally craps out performance of very large files on
> most pnfs-objects layouts, because of how the map changes when the
> file has grown and spills into the next raid group.
> 
> CC: Stable Tree <[e-mail]>
> Signed-off-by: Boaz Harrosh <[e-mail]>

Trond hi

I'm sitting on this bug for over 6 month now. I completely forgot about
it until QA moved to new Fedora and a vanila Kernel which is missing this
fix.

This is a real bummer for objects, in the case of large clusters for example
a Panasas cluster with two shelves and up. So on big clusters where performance
should be better, but with out this fix it is miserably unacceptedly slow.

Thanks
Boaz

--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index d53d678..3264fca 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -7058,7 +7058,7 @@  static void nfs4_layoutget_done(struct rpc_task *task, void *calldata)
 	struct nfs4_state *state = NULL;
 	unsigned long timeo, giveup;
 
-	dprintk("--> %s\n", __func__);
+	dprintk("--> %s tk_status => %d\n", __func__, task->tk_status);
 
 	if (!nfs41_sequence_done(task, &lgp->res.seq_res))
 		goto out;
@@ -7067,11 +7067,27 @@  static void nfs4_layoutget_done(struct rpc_task *task, void *calldata)
 	case 0:
 		goto out;
 	case -NFS4ERR_LAYOUTTRYLATER:
+	/* NFS4ERR_RECALLCONFLICT is always a minimal delay (conflict with
+	 * self)
+	 * TODO: NFS4ERR_LAYOUTTRYLATER is a conflict with another client
+	 * (or clients). What we should do is randomize a short delay like on a
+	 * network broadcast burst, and raise the random max every failure.
+	 * For now leave it stateless and do this polling.
+	 */
 	case -NFS4ERR_RECALLCONFLICT:
 		timeo = rpc_get_timeout(task->tk_client);
 		giveup = lgp->args.timestamp + timeo;
-		if (time_after(giveup, jiffies))
-			task->tk_status = -NFS4ERR_DELAY;
+		if (time_after(giveup, jiffies)) {
+			/* Do a minimum delay, We are actually waiting for our
+			 * own IO to finish (In most cases)
+			 */
+			dprintk("%s: NFS4ERR_RECALLCONFLICT waiting\n",
+				__func__);
+			rpc_delay(task, NFS4_POLL_RETRY_MIN);
+			task->tk_status = 0;
+			rpc_restart_call_prepare(task);
+			goto out; /* Do not call nfs4_async_handle_error() */
+		}
 		break;
 	case -NFS4ERR_EXPIRED:
 	case -NFS4ERR_BAD_STATEID: