sin/samba-mirror
1
0
Fork 0
mirror of https://github.com/samba-team/samba.git synced 2024-12-23 17:34:34 +03:00
Code

ctdb-ipalloc: Fix cumulative takeover timeout

Browse Source 
Commit c40fc62642 runs the IP allocation
algorithm after calculating the timeout offset.  If the algorithm
takes a long time then there may be no attempt to release or take over
IPs.

Instead, reset the timeout just before the RELEASE_IP stage if an
early jump to IPREALLOCATED was not taken.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12161

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

Autobuild-User(master): Amitay Isaacs <amitay@samba.org>
Autobuild-Date(master): Thu Aug 18 12:36:37 CEST 2016 on sn-devel-144
This commit is contained in:
Martin Schwenke 2016-08-18 12:57:33 +10:00 committed by Amitay Isaacs
parent b21308252e
commit 626dcc9e49
1 changed files with 17 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
ctdb/server/ctdb_takeover.c
View File

@ -1539,16 +1539,9 @@ int ctdb_takeover_run(struct ctdb_context *ctdb, struct ctdb_node_map_old *nodem
return -1;
}
/* Each of the later stages (RELEASE_IP, TAKEOVER_IP,
* IPREALLOCATED) notionally has a timeout of TakeoverTimeout
* seconds. However, RELEASE_IP can take longer due to TCP
* connection killing, so sometimes needs more time.
* Therefore, use a cumulative timeout of TakeoverTimeout * 3
* seconds across all 3 stages. No explicit expiry checks are
* needed before each stage because tevent is smart enough to
* fire the timeouts even if they are in the past. Initialise
* this here to cope with early jumps to IPREALLOCATED. */
timeout = timeval_current_ofs(3 * ctdb->tunable.takeover_timeout,0);
/* Default timeout for early jump to IPREALLOCATED. See below
* for explanation of 3 times... */
timeout = timeval_current_ofs(3 * ctdb->tunable.takeover_timeout, 0);
/*
* ip failover is completely disabled, just send out the
@ -1622,6 +1615,20 @@ int ctdb_takeover_run(struct ctdb_context *ctdb, struct ctdb_node_map_old *nodem
ZERO_STRUCT(ip); /* Avoid valgrind warnings for union */
/* Each of the following stages (RELEASE_IP, TAKEOVER_IP,
* IPREALLOCATED) notionally has a timeout of TakeoverTimeout
* seconds. However, RELEASE_IP can take longer due to TCP
* connection killing, so sometimes needs more time.
* Therefore, use a cumulative timeout of TakeoverTimeout * 3
* seconds across all 3 stages. No explicit expiry checks are
* needed before each stage because tevent is smart enough to
* fire the timeouts even if they are in the past. Initialise
* this here so it explicitly covers the stages we're
* interested in but, in particular, not the time taken by the
* ipalloc().
*/
timeout = timeval_current_ofs(3 * ctdb->tunable.takeover_timeout, 0);
/* Send a RELEASE_IP to all nodes that should not be hosting
* each IP. For each IP, all but one of these will be
* redundant. However, the redundant ones are used to tell