Description of problem: Upgrade to firefox-10.0.4-1.el5_8-x86_64 from firefox-3.0.6-1.el5.x86_64.rpm. is found that firefox 10.0.4 hangs when loading most web pages on glusterfs. Please find attached the output of: strace -s 1024 -v /usr/bin/firefox The hang was induced by browsing to google.com The process was sent a SIGINT to terminate. The server is headless, the process was using a remote X11 server for DISPLAY. Version-Release number of selected component (if applicable): How reproducible: Always Steps to Reproduce: GlusterFS version 3.2.6, RHEL5.8 Client x86_64 - install firefox-10.0.4-1 run it over GlusterFS mount, open a webpage and just hangs. Actual results: Firefox hang Expected results: Firefox renders web page successfully.
Created attachment 597878 [details] strace firefox
Created attachment 597879 [details] Gluster dump during firefox hang
Created attachment 597880 [details] Sosreport
Waiting on * tar czf gluster_configuration.tar.gz /etc/glusterd from servers * tar czf gluster_logs_servers.tar.gz /var/log/glusterfs from servers * tar czf gluster_logs_client.tar.gz /var/log/glusterfs from client.
Created attachment 597944 [details] GlusterFS config melnode1 The two servers are also the clients, there are no other clients, so the log archives contain both the client and server process output.
Created attachment 597945 [details] GlusterFS config melnode2
Created attachment 597947 [details] Gluster Client/Server logs melnode1
Created attachment 597948 [details] Gluster Client/Server logs melnode2
Harsha, I will be going through the logs now, meantime, wanted to understand what would be the behavior on 3.3.0 release? Is it possible to test that here? -Amar
[global.callpool.stack.1.frame.2] global.callpool.stack.1.frame.2.ref_count=0 global.callpool.stack.1.frame.2.translator=u0-client-0 global.callpool.stack.1.frame.2.complete=0 global.callpool.stack.1.frame.2.parent=u0-replicate-0 global.callpool.stack.1.frame.2.wind_from=afr_lk global.callpool.stack.1.frame.2.wind_to=priv->children[i]->fops->lk global.callpool.stack.1.frame.2.unwind_to=afr_lk_cbk Seems some 'missing' frame in client-protocol. Let me get more info.
Created attachment 602002 [details] Gluster statedump 3.3.0 http://pastebin.gsslab.fab.redhat.com/mypaste/bLHk/ - testing with RHS 2.0 and firefox 10.0.4
Created attachment 602003 [details] Gluster statedump 3.3.0
Created attachment 602060 [details] Sosreport node1
Created attachment 602061 [details] Sosreport node2
Amar, Testing with RHS 2.0 / GlusterFS 3.3.0 succeeds with no infinite hangs as it would with 3.2.x, but the real problem is that now the browser waits for like 3mins to open up a 'web' link. You could see a .parentlock file where the firefox is issuing a WRITE locking request (F_WRLCK) Also odd thing is that a lot of files are being cached by io-cache for some reason, could that be cause for the slowness? Do you have any inputs?
http://rhn.redhat.com/errata/RHSA-2012-1350.html - believed to be a bug in Firefox , storage.nfs_filesystem not being enabled. Upstream mozilla relevant bug - https://bugzilla.mozilla.org/show_bug.cgi?id=433129