Skip to contentSkip to conversation

dlm: Don't send callback to node making lock request when "try 1cb" fails

PATCH Steven Whitehouse 2 participants 2 messages Sep 3, 2010 Distributed lock manager (DLM)

lore
Share & tools
lore
#0 lore raw

When converting a lock, an lkb is in the granted state and also being used to request a new state. In the case that the conversion was a "try 1cb" type which has failed, and if the new state was incompatible with the old state, a callback was being generated to the requesting node. This is incorrect as callbacks should only be sent to all the other nodes holding blocking locks. The requesting node should receive the normal (failed) response to its "try 1cb" conversion request only.

This was discovered while debugging a performance problem on GFS2, however this fix also speeds up GFS as well. In the GFS2 case the performance gain is over 10x for cases of write activity to an inode whose glock is cached on another, idle (wrt that glock) node.

fs/dlm/lock.c +2 -0
index 031dbe3..dae42e1 100644
@@ -1846,6 +1846,8 @@ static void send_bast_queue(struct dlm_rsb *r, struct list_head *head,
 	struct dlm_lkb *gr;
 
 	list_for_each_entry(gr, head, lkb_statequeue) {
+		if (gr == lkb)
+			continue;
 		if (gr->lkb_bastfn && modes_require_bast(gr, lkb)) {
 			queue_bast(r, gr, lkb->lkb_rqmode);
 			gr->lkb_highbast = lkb->lkb_rqmode;
maintainer Tested-by Abhijith Das
#1 lore raw
Quoted 9 lines · Steven Whitehouse

When converting a lock, an lkb is in the granted state and also being used to request a new state. In the case that the conversion was a "try 1cb" type which has failed, and if the new state was incompatible with the old state, a callback was being generated to the requesting node. This is incorrect as callbacks should only be sent to all the other nodes holding blocking locks. The requesting node should receive the normal (failed) response to its "try 1cb" conversion request only.

Quoted 4 lines

This was discovered while debugging a performance problem on GFS2, however this fix also speeds up GFS as well. In the GFS2 case the performance gain is over 10x for cases of write activity to an inode whose glock is cached on another, idle (wrt that glock) node.

Thanks, I added a comment and queued it up.
Dave

From 314dd2a05340fffbe217c5e40ec6c3bd1d07bf89 Mon Sep 17 00:00:00 2001
From: Steven Whitehouse <[email protected]>
Date: Fri, 3 Sep 2010 10:07:48 -0500
Subject: [PATCH] dlm: Don't send callback to node making lock request when "try 1cb" fails

When converting a lock, an lkb is in the granted state and also being used to request a new state. In the case that the conversion was a "try 1cb" type which has failed, and if the new state was incompatible with the old state, a callback was being generated to the requesting node. This is incorrect as callbacks should only be sent to all the other nodes holding blocking locks. The requesting node should receive the normal (failed) response to its "try 1cb" conversion request only.

This was discovered while debugging a performance problem on GFS2, however this fix also speeds up GFS as well. In the GFS2 case the performance gain is over 10x for cases of write activity to an inode whose glock is cached on another, idle (wrt that glock) node.

(comment added, dct)

---
 fs/dlm/lock.c |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)
fs/dlm/lock.c +3 -0
index 031dbe3..64e5f3e 100644
@@ -1846,6 +1846,9 @@ static void send_bast_queue(struct dlm_rsb *r, struct list_head *head,
 	struct dlm_lkb *gr;
 
 	list_for_each_entry(gr, head, lkb_statequeue) {
+		/* skip self when sending basts to convertqueue */
+		if (gr == lkb)
+			continue;
 		if (gr->lkb_bastfn && modes_require_bast(gr, lkb)) {
 			queue_bast(r, gr, lkb->lkb_rqmode);
 			gr->lkb_highbast = lkb->lkb_rqmode;
Signature · 1 line
1.7.1.1