[
https://jira.jboss.org/browse/ISPN-715?page=com.atlassian.jira.plugin.sys...
]
Mircea Markus commented on ISPN-715:
------------------------------------
thinking about it not sure how that would reduce number of deadlocks?
[14:57] mmarkus: or cause deadlocks
[14:58] an1310: ?
[14:59] an1310: You would only acquire the one lock. If node A and node B both acquired a
local lock, but we hashed to node C, wouldn't that be a deadlock?
[15:01] mmarkus: so we have tx1 running on A, tx2 running on B and both work on a key
that's hashed to C
[15:01] mmarkus: is this your scenario?
[15:03] an1310: Yes, unless there's some way that I could tell my app to always go to
C (since that's where the data lives).
[15:04] an1310: I'm colocating all the necessary caches involved in a tx on one data
node and using that "eager lock single node" feature.
[15:04] an1310: ...data on a cache; typing too fast.
[15:04] mmarkus: returning to the example: how does that cause a deadlock?
[15:05] mmarkus: i.e. the fact that I also acquire a local lock on A and B for key
"k" which is hashed to C
[15:06] an1310: I may be wrong about the deadlock part.
[15:08] mmarkus: I don't see the benefit of it. Actually I think it would even cause
performance faults: if another tx3 runs on A and tries to acquire lock on "k"
which is held by tx1 then it would have to go to C and only then to timeout. Right now it
would fail fast, without need to go to C
Eager locking a single node should only acquire a single lock
-------------------------------------------------------------
Key: ISPN-715
URL:
https://jira.jboss.org/browse/ISPN-715
Project: Infinispan
Issue Type: Feature Request
Reporter: Erik Salter
Assignee: Manik Surtani
The new eager/explicit lock for a single node feature in 4.2.0 can be further optimized.
Right now, a local lock and a lock on the node where the key is hashed is acquired. To
increase throughput and minimize deadlock occurrences, we should only acquire the one lock
(on the node).
(Obviously, if the key is hashed to the local node, the local lock and the node lock are
one in the same).
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
https://jira.jboss.org/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira