[ https://issues.apache.org/jira/browse/TS-1590?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13669798#comment-13669798 ]
Leif Hedstrom commented on TS-1590: ----------------------------------- I suspect a refactoring of remap processor is required, to allow it to schedule the appropriate events back to the calling ET_NET (net-thread). There's no reason why the do_http_server_open() should have to happen on a remap thread. You are definitely spot on, there is no l1_hash on the remap processor threads... A real ugly solution would be to add that to them, but a better solution is to change the remap processor to handle the continuation back to an Net-thread before it needs the l1_hash. I'm going to move this out to v3.5.0, unless I hear otherwise. I think it's somewhat reasonable to say that you have to set share_server_sessions to 0 or 1 if you really have to use the remap processor (which I don't know of anyone using). > use_remap_processor crashes if share_server_sessions = 2 > -------------------------------------------------------- > > Key: TS-1590 > URL: https://issues.apache.org/jira/browse/TS-1590 > Project: Traffic Server > Issue Type: Bug > Components: HTTP > Affects Versions: 3.3.0 > Reporter: Conan Wang > Priority: Critical > Fix For: 3.3.3 > > > easy to reproduce: > {code} > CONFIG proxy.config.remap.use_remap_processor INT 1 (default is 0) > CONFIG proxy.config.http.share_server_sessions INT 2 (0, 1 will be ok) > {code} > {code} > Program received signal EXC_BAD_ACCESS, Could not access memory. > Reason: KERN_INVALID_ADDRESS at address: 0x0000000000001210 > [Switching to process 8927 thread 0x1a03] > 0x00000001000ead15 in _acquire_session (bucket=0x11c0, ip=0x6713138, > hostname_hash=@0xb06915d0, sm=0x6712aa0) at HttpSessionManager.cc:191 > (gdb) bt > #0 0x00000001000ead15 in _acquire_session (bucket=0x11c0, ip=0x6713138, > hostname_hash=@0xb06915d0, sm=0x6712aa0) at HttpSessionManager.cc:191 > #1 0x00000001000eb366 in HttpSessionManager::acquire_session > (this=0x100445e60, cont=0x6712aa0, ip=0x6713138, hostname=0x4ebc19 > "127.0.0.1", ua_session=0x2a88980, sm=0x6712aa0) at HttpSessionManager.cc:274 > #2 0x0000000100105c29 in HttpSM::do_http_server_open (this=0x6712aa0, > raw=false) at HttpSM.cc:4384 > ...... > (gdb) p this_ethread()->l1_hash > $2 = (SessionBucket *) 0x0 > (gdb) p this_ethread()->event_types > $3 = 2 (ET_REMAP) > {code} > Using separate remap processor is a hidden option, and I enable it by > accident.. (Does anyone use it in prod?) > I noticed HttpSM::do_http_server_open is always executed by the remap > processer ethread (because of > action.continuation->handleEvent(EVENT_REMAP_COMPLETE, NULL), correct me if > wrong). While the remap thread was not initialized as ET_NET and has no > l1_hash, server session lookup in this ET_REMAP thread will crash. > I didn't find a quick way to make HttpSM handling EVENT_REMAP_COMPLETE on > ET_NET. So a fast workaround would be falling back to global server > connection when use_remap_processor. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira