couchdb-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Adam Kocoloski (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (COUCHDB-1346) CouchDB hangs during start of view indexing
Date Sun, 02 Dec 2012 14:03:58 GMT

    [ https://issues.apache.org/jira/browse/COUCHDB-1346?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13508271#comment-13508271
] 

Adam Kocoloski commented on COUCHDB-1346:
-----------------------------------------

I did some more reading and log-based debugging this morning.  We're doing something rather
screwy.  It turns out the the individual view index processes are *already* monitoring their
parent DB; when we delete the DB, the index is automatically closed:

https://github.com/apache/couchdb/blob/bde29b/src/couch_index/src/couch_index.erl#L80
https://github.com/apache/couchdb/blob/bde29b/src/couch_index/src/couch_index.erl#L296-L300

The handler for that 'DOWN' message will invoke the gen_server's terminate function, which
closes the file descriptor cleanly.  So that's A Good Thing.  Unfortunately, we don't always
let that cleanup run to completion, because we've got this separate DB update notification
listener:

https://github.com/apache/couchdb/blob/bde29b/src/couch_index/src/couch_index_server.erl#L159-L168

The couch_index processes do not trap exits, so when the reset_indexes function calls shutdown_sync
it terminates the couch_index process immediately, bypassing any additional cleanup that we
wanted to do.  The patch I wrote allows for the termination to finish cleanly.

It seems to me that even with the existing shutdown_sync invocation we'd eventually close
all the file descriptors because of exit signal propagation, but at that point we may be racing
the process that tries to delete them (and when we lose, we hang).  The clean shutdown avoids
that race.
                
> CouchDB hangs during start of view indexing
> -------------------------------------------
>
>                 Key: COUCHDB-1346
>                 URL: https://issues.apache.org/jira/browse/COUCHDB-1346
>             Project: CouchDB
>          Issue Type: Bug
>          Components: View Server Support
>    Affects Versions: 1.3
>         Environment: Windows 7 Enterprise only, not able to replicate on Mac OS X.
> Erlang R14B03 + crypto patches.
> Mozilla Javascript 1.8.5
>            Reporter: Dave Cottlehuber
>            Assignee: Adam Kocoloski
>            Priority: Blocker
>              Labels: Windows
>             Fix For: 1.3
>
>
> [info] [<0.20499.0>] Opening index for db: test_suite_db idx: f4421bf4e9c9bf2acb3db91bca9e9adc
sig: "d5c87ad33242b181f86be2139cbccd96"
> [info] [<0.20504.0>] Starting index update for db: test_suite_db idx: f4421bf4e9c9bf2acb3db91bca9e9adc
> [info] [<0.20334.0>] 172.16.40.1 - - POST /test_suite_db/_temp_view 500
> [info] [<0.20513.0>] 172.16.40.1 - - GET /_utils/couch_tests.html?script/couch_tests.js
200
> [info] [<0.20514.0>] 172.16.40.1 - - GET /_utils/index.html 200
> [info] [<0.20060.0>] 172.16.40.1 - - DELETE /test_suite_db_a/ 200
> [info] [<0.20407.0>] 172.16.40.1 - - GET /test_suite_reports/ 404
> [info] [<0.20058.0>] 172.16.40.1 - - DELETE /test_suite_db/ 404
> [info] [<0.20071.0>] 172.16.40.1 - - DELETE /test_suite_db/ 404
> [info] [<0.20069.0>] 172.16.40.1 - - DELETE /test_suite_db/ 404
> [info] [<0.20484.0>] 172.16.40.1 - - DELETE /test_suite_db/ 404
> [info] [<0.20364.0>] 172.16.40.1 - - DELETE /test_suite_db/ 404
> [info] [<0.20062.0>] 172.16.40.1 - - DELETE /test_suite_db/ 404
> [info] [<0.20388.0>] 172.16.40.1 - - DELETE /test_suite_db/ 404
> [info] [<0.20345.0>] 172.16.40.1 - - DELETE /test_suite_db/ 404
> [info] [<0.20072.0>] 172.16.40.1 - - DELETE /test_suite_db/ 404
> [info] [<0.20059.0>] 172.16.40.1 - - DELETE /test_suite_db/ 404
> [info] [<0.20061.0>] 172.16.40.1 - - DELETE /test_suite_db/ 404
> [info] [<0.20472.0>] 172.16.40.1 - - DELETE /test_suite_db/ 200
> [error] [<0.20050.0>] ** Generic server couch_index_server terminating 
> ** Last message in was {'$gen_cast',{reset_indexes,<<"test_suite_db">>}}
> ** When Server state == {st,"../var/lib/couchdb"}
> ** Reason for termination == 
> ** {{case_clause,{error,eacces}},
>     [{couch_file,'-nuke_dir/2-fun-0-',3},
>      {lists,foreach,2},
>      {couch_file,nuke_dir,2},
>      {couch_index_server,handle_cast,2},
>      {gen_server,handle_msg,5},
>      {proc_lib,init_p_do_apply,3}]}
> =ERROR REPORT==== 23-Nov-2011::21:17:14 ===
> ** Generic server couch_index_server terminating 
> ** Last message in was {'$gen_cast',{reset_indexes,<<"test_suite_db">>}}
> ** When Server state == {st,"../var/lib/couchdb"}
> ** Reason for termination == 
> ** {{case_clause,{error,eacces}},
>     [{couch_file,'-nuke_dir/2-fun-0-',3},
>      {lists,foreach,2},
>      {couch_file,nuke_dir,2},
>      {couch_index_server,handle_cast,2},
>      {gen_server,handle_msg,5},
>      {proc_lib,init_p_do_apply,3}]}
> [error] [<0.20050.0>] {error_report,<0.19957.0>,
>                           {<0.20050.0>,crash_report,
>                            [[{initial_call,
>                                  {couch_index_server,init,['Argument__1']}},
>                              {pid,<0.20050.0>},
>                              {registered_name,couch_index_server},
>                              {error_info,
>                                  {exit,
>                                      {{case_clause,{error,eacces}},
>                                       [{couch_file,'-nuke_dir/2-fun-0-',3},
>                                        {lists,foreach,2},
>                                        {couch_file,nuke_dir,2},
>                                        {couch_index_server,handle_cast,2},
>                                        {gen_server,handle_msg,5},
>                                        {proc_lib,init_p_do_apply,3}]},
>                                      [{gen_server,terminate,6},
>                                       {proc_lib,init_p_do_apply,3}]}},
>                              {ancestors,
>                                  [couch_secondary_services,couch_server_sup,
>                                   <0.19958.0>]},
>                              {messages,
>                                  [{'$gen_cast',
>                                       {reset_indexes,<<"test_suite_db_a">>}}]},
>                              {links,[<0.20051.0>,<0.20026.0>]},
>                              {dictionary,[]},
>                              {trap_exit,true},
>                              {status,running},
>                              {heap_size,1597},
>                              {stack_size,24},
>                              {reductions,12211}],
>                             [{neighbour,
>                                  [{pid,<0.20051.0>},
>                                   {registered_name,[]},
>                                   {initial_call,
>                                       {couch_event_sup,init,['Argument__1']}},
>                                   {current_function,{gen_server,loop,6}},
>                                   {ancestors,
>                                       [couch_index_server,
>                                        couch_secondary_services,
>                                        couch_server_sup,<0.19958.0>]},
>                                   {messages,[]},
>                                   {links,[<0.20050.0>,<0.20018.0>]},
>                                   {dictionary,[]},
>                                   {trap_exit,false},
>                                   {status,waiting},
>                                   {heap_size,233},
>                                   {stack_size,9},
>                                   {reductions,32}]}]]}}
> =CRASH REPORT==== 23-Nov-2011::21:17:14 ===
>   crasher:
>     initial call: couch_index_server:init/1
>     pid: <0.20050.0>
>     registered_name: couch_index_server
>     exception exit: {{case_clause,{error,eacces}},
>                      [{couch_file,'-nuke_dir/2-fun-0-',3},
>                       {lists,foreach,2},
>                       {couch_file,nuke_dir,2},
>                       {couch_index_server,handle_cast,2},
>                       {gen_server,handle_msg,5},
>                       {proc_lib,init_p_do_apply,3}]}
>       in function  gen_server:terminate/6
>     ancestors: [couch_secondary_services,couch_server_sup,<0.19958.0>]
>     messages: [{'$gen_cast',{reset_indexes,<<"test_suite_db_a">>}}]
>     links: [<0.20051.0>,<0.20026.0>]
>     dictionary: []
>     trap_exit: true
>     status: running
>     heap_size: 1597
>     stack_size: 24
>     reductions: 12211
>   neighbours:
>     neighbour: [{pid,<0.20051.0>},
>                   {registered_name,[]},
>                   {initial_call,{couch_event_sup,init,['Argument__1']}},
>                   {current_function,{gen_server,loop,6}},
>                   {ancestors,[couch_index_server,couch_secondary_services,
>                               couch_server_sup,<0.19958.0>]},
>                   {messages,[]},
>                   {links,[<0.20050.0>,<0.20018.0>]},
>                   {dictionary,[]},
>                   {trap_exit,false},
>                   {status,waiting},
>                   {heap_size,233},
>                   {stack_size,9},
>                   {reductions,32}]
> [error] [<0.20026.0>] {error_report,<0.19957.0>,
>                           {<0.20026.0>,supervisor_report,
>                            [{supervisor,{local,couch_secondary_services}},
>                             {errorContext,child_terminated},
>                             {reason,
>                                 {{case_clause,{error,eacces}},
>                                  [{couch_file,'-nuke_dir/2-fun-0-',3},
>                                   {lists,foreach,2},
>                                   {couch_file,nuke_dir,2},
>                                   {couch_index_server,handle_cast,2},
>                                   {gen_server,handle_msg,5},
>                                   {proc_lib,init_p_do_apply,3}]}},
>                             {offender,
>                                 [{pid,<0.20050.0>},
>                                  {name,index_server},
>                                  {mfargs,{couch_index_server,start_link,[]}},
>                                  {restart_type,permanent},
>                                  {shutdown,brutal_kill},
>                                  {child_type,worker}]}]}}
> OS process tree at this time is:
> Process information for SENDAI:
> Name                             Pid Pri Thd  Hnd      VM      WS    Priv
> Idle                               0   0   2    0       0      24       0
>   System                           4   8  79  477    3380     304     108
> explorer                        1984   8  21  664  213732   46340   21540
>   cmd                           2104   8   1   25   48132    3304    2144
>     pslist                      2776  13   1  133   63584    4976    2000
>   cmd                           2504   8   1   26   44980    3512    3012
>     werl                        2680   8  16  390  196232   40064   28628
>       win32sysinfo              1152   8   1   21   12624    2124     640
>       couchspawnkillable        1444   8   1   30   12992    2284     688
>         couchjs                 1468   8   1   39   55900    6572    4056
>       couchspawnkillable        2740   8   1   30   12992    2280     684
>         couchjs                 2756   8   1   39   55900    7108    4444
> Erlang resumes running CouchDB when couchjs procs are terminated with extreme
> prejudice. The hang still occurs after reverting fdmanana's COUCHDB-1334
> commit. This could be a race condition during invalidation of the views, and
> subsequent deletion of the related ddoc view directory prior to reindexing.
> On Windows a filesystem object cannot be deleted if there are open handles
> remaining.

--
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

Mime
View raw message