| Bug #37334 | Very rare race-condition can lead to stale-connection | ||
|---|---|---|---|
| Submitted: | 11 Jun 2008 9:22 | Modified: | 2 Jul 2008 15:52 |
| Reporter: | Jonas Oreland | Email Updates: | |
| Status: | Closed | Impact on me: | |
| Category: | MySQL Cluster: Cluster (NDB) storage engine | Severity: | S3 (Non-critical) |
| Version: | * | OS: | Any |
| Assigned to: | Jonas Oreland | CPU Architecture: | Any |
[11 Jun 2008 9:28]
Bugs System
A patch for this bug has been committed. After review, it may be pushed to the relevant source trees for release in the next version. You can access the patch from: http://lists.mysql.com/commits/47722 2621 jonas@mysql.com 2008-06-11 ndb - bug#37334 Fix race condition with sequence do_connect,do_disconnect on transporter
[2 Jul 2008 6:58]
Jonas Oreland
pushed to 6.3.16 and 6.2.16
[2 Jul 2008 15:52]
Jon Stephens
Documented in the NDB 6.2.16 and 6.3.16 changelogs as follows:
In rare circumstances, a connection followed by a disconnection could give
rise to a "stale" connection where the connection still existed but was
not seen by the transporter.
[12 Dec 2008 23:27]
Bugs System
Pushed into 6.0.6-alpha (revid:jonas@mysql.com-20080611092840-7840su31p7ks71u9) (version source revid:jonas@mysql.com-20080812185642-1nevjb94zj621dqx) (pib:5)
Description: 1) do_connect state := CONNECTING 2) start client_thread reads connectState and starts connecting 3) do_disconnect state := DISCONNECTING 4) update connections state == DISCONNECTING && isConnected() == false -> report_disconnect -> state := DISCONNECTED 5) start_client_thread manage to connect --- End result: state = DISCONNECTED transporter is connected How to repeat: no repeatable testcase :-( Suggested fix: + case DISCONNECTED: + { + if (t->isConnected()) + { + g_eventLogger.warning("Found connection to %u in state DISCONNECTED " + " while being connected, disconnecting!", + t->getRemoteNodeId()); + t->doDisconnect(); + } + break; + }