From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 16923 invoked by alias); 16 Apr 2008 14:32:37 -0000 Received: (qmail 16889 invoked by uid 9453); 16 Apr 2008 14:32:36 -0000 Date: Wed, 16 Apr 2008 14:32:00 -0000 Message-ID: <20080416143236.16872.qmail@sourceware.org> From: teigland@sourceware.org To: cluster-cvs@sources.redhat.com, cluster-devel@redhat.com Subject: Cluster Project branch, master, updated. gfs-kernel_0_1_22-180-ga187481 X-Git-Refname: refs/heads/master X-Git-Reftype: branch X-Git-Oldrev: 257d1b44ad25884c05606773f7a48898b13aed55 X-Git-Newrev: a187481779ec3a5321c92a9f9ae587dcacd64d48 Mailing-List: contact cluster-cvs-help@sourceware.org; run by ezmlm Precedence: bulk List-Id: List-Subscribe: List-Post: List-Help: , Sender: cluster-cvs-owner@sourceware.org X-SW-Source: 2008-q2/txt/msg00118.txt.bz2 This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "Cluster Project". http://sources.redhat.com/git/gitweb.cgi?p=cluster.git;a=commitdiff;h=a187481779ec3a5321c92a9f9ae587dcacd64d48 The branch, master has been updated via a187481779ec3a5321c92a9f9ae587dcacd64d48 (commit) from 257d1b44ad25884c05606773f7a48898b13aed55 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- commit a187481779ec3a5321c92a9f9ae587dcacd64d48 Author: David Teigland Date: Wed Apr 16 09:22:27 2008 -0500 gfs_controld: retry recovery for withdrawn journal bz 442451 This is unfortunate, but seems to be the best solution available. The problem, described more fully in the bz, is that when gfs_controld tries to do recovery on a journal for a withdraw, the withdrawing node may not yet have cleared its dlm locks. This means the journal lock may still be held by the withdrawing node, causing all the recovering node(s) to fail acquiring it, and no one does the recovery. The solution is for all recovering nodes to retry recovery of a withdrawn journal until they succeed (only the first to get the journal lock will actually recover it, the others will see it's recovered and report success.) Signed-off-by: David Teigland ----------------------------------------------------------------------- Summary of changes: group/gfs_controld/recover.c | 19 +++++++++++++++++++ 1 files changed, 19 insertions(+), 0 deletions(-) diff --git a/group/gfs_controld/recover.c b/group/gfs_controld/recover.c index 9ce3aa7..52d96ff 100644 --- a/group/gfs_controld/recover.c +++ b/group/gfs_controld/recover.c @@ -1913,6 +1913,25 @@ int kernel_recovery_done(char *table) switch (atoi(buf)) { case LM_RD_GAVEUP: + /* + * This is unfortunate; it's needed for bz 442451 where + * gfs-kernel fails to acquire the journal lock on all nodes + * because a withdrawing node has not yet called + * dlm_release_lockspace() to free it's journal lock. With + * this, all nodes should repeatedly try to to recover the + * journal of the withdrawn node until the withdrawing node + * clears its dlm locks, and gfs on each of the remaining nodes + * succeeds in doing the recovery. + */ + + if (memb->withdrawing) { + log_group(mg, "recovery_done jid %d nodeid %d retry " + "for withdraw", memb->jid, memb->nodeid); + memb->tell_gfs_to_recover = 1; + memb->wait_gfs_recover_done = 0; + usleep(500000); + } + memb->local_recovery_status = RS_GAVEUP; ss = "gaveup"; break; hooks/post-receive -- Cluster Project