From nobody@hyperreal.com Tue Apr 1 23:35:43 1997 Received: by taz.hyperreal.com (8.8.4/V2.0) id XAA28294; Tue, 1 Apr 1997 23:35:43 -0800 (PST) Message-Id: <199704020735.XAA28294@taz.hyperreal.com> Date: Tue, 1 Apr 1997 23:35:43 -0800 (PST) From: Steiner@hyperreal.com, Hugo Reply-To: hugo@hunterlink.net.au To: apbugs@hyperreal.com Subject: server stops serving for a period and then resumes with no messages X-Send-Pr-Version: 3.2 >Number: 299 >Category: config >Synopsis: server stops serving for a period and then resumes with no messages >Confidential: no >Severity: critical >Priority: medium >Responsible: apache >State: closed >Class: duplicate >Submitter-Id: apache >Arrival-Date: Tue Apr 1 23:40:03 1997 >Last-Modified: Sun Apr 27 14:51:58 PDT 1997 >Originator: hugo@hunterlink.net.au >Organization: >Release: 1.2b7 >Environment: OS: Digital Unix 4.0B, no patches. Compiler: gcc >Description: The web server stops serving documents. Connections continue to be accepted, however, no data is returned. This typically lasts about 20 minutes and then goes away. Killing a child or sending a SIGHUP will cause the server to resume (Signal?) The parent is typically in a wait call. The same problem was experienced with 1.1.3 This resembles PR256 which was reported on BSDi In addition to this, we seem to have hit a limit in virtual servers of about 22 when more than this are confiugred up, you get a 'Can't bind to xxx.xxx.xxx.xxx' on the last VIP in the http.conf file. This also happens with 1.1.3. Chnaging the memory allocation size from 8k to 16k increases the number of servers that can be configured. >How-To-Repeat: Unsure. Generally it seems to happen when the server has been up for more than 6 hours. however it has also happened a couple of times within this period. >Fix: Chnaging the memory allocation size from 8k to 16k increases the number of servers that can be configured. Unfortunatley, no other pointers are availible >Audit-Trail: State-Changed-From-To: open-feedback State-Changed-By: coar@decus.org State-Changed-When: Wed Apr 23 08:27:23 PDT 1997 State-Changed-Why: There were some fixes made in the 1.2b8 beta that probably affect the "can't bind" message. Can you try upgrading to that version and see whether the problem persists? Also, please check the FAQ on the subject of the number of virtual hosts (). Please let us know what you find.. Category-Changed-From-To: other-config Category-Changed-By: coar@decus.org Category-Changed-When: Wed Apr 23 08:27:23 PDT 1997 State-Changed-From-To: feedback-analyzed State-Changed-By: coar@decus.org State-Changed-When: Thu Apr 24 03:20:50 PDT 1997 State-Changed-Why: The customer reports: "1.2B8 seems to have fixed this problem. It was a drop in and work where we had to hack at 1.2b7 to get it going. The stalling seems to have been fixed by adding -DUSE_FCNTL_SERIALISED_ACCEPT but we have not run it long enough to be sure yet." Thank you for the update; please let us know your conclusion wrt the USE_FCNTL_SERIALISED_ACCEPT. State-Changed-From-To: analyzed-closed State-Changed-By: dgaudet State-Changed-When: Sun Apr 27 14:51:57 PDT 1997 State-Changed-Why: Superceded by PR#467 which talks generally of the problem with multiple Listens and serialized accept. Thanks for using Apache! Dean >Unformatted: