MantisBT - Issues

0001082: searchd won't shut down with threads workers

Sat, 04 Feb 2012 02:58:44 +0100

I've documented a lot of this on the forums:
http://sphinxsearch.com/forum/view.html?id=8971 [^]

Further investigation has found the following: if I turn seamless_rotate off, then the first rotate attempted hangs. Otherwise, with seamless_rotate on, starting the daemon and stopping it straight away is fine. Starting, running a query, and then stopping causes the stopping to hang.

I've added a few info debugging lines into the 2.0.3-release source, and it seems to be stuck when telling the rotation thread to shut down (line 1278 of src/searchd.cpp). It certainly doesn't get to the point of a few lines down with "int64_t tmShutStarted = sphMicroTimer();".

0001081: group-count-distinct error on select with two indexes -- Sphinx 2.0.3-id64-release (r3043)

Fri, 03 Feb 2012 12:53:51 +0100

We have
trainings - main index
trainings_delta - delta index
with same set of attributes

Error in query.log

/* Fri Feb 3 13:25:41.666 2012 conn 110 wall 0.000 found 2 */ SELECT company_id, COUNT(DISTINCT training_id) FROM trainings, trainings_delta WHERE is_active=1 company_id IN (88,3034,3084,3087,3101) GROUP BY company_id; # error=group-count-distinct attribute 'training_id' not found

0000885: Bitwise operators doesn't work

Thu, 02 Feb 2012 18:50:43 +0100

Can't use bitwise operator in setSelect in php api.

source code:

require_once(dirname(__FILE__)."/sphinx-2.0.1-beta/api/sphinxapi.php");
$cl = new SphinxClient();
$cl->setServer('localhost', 3312);
$cl->setSelect("outlet_id&8 as new_id");
$response = $cl->query('','newmediadb');
echo $cl->getLastError());
var_dump($response);

output:

searchd error: select: syntax error, unexpected $undefined, expecting SEL_TOKEN near '&8 as new_id'

bool(false)

outlet_id is int attribute

0001070: random weight in SPH_RANK_EXPR in 2.0.3

Thu, 02 Feb 2012 13:26:13 +0100

Ok actually im trying new ranker system on 2.0.3
And i meet very strange behaviour:
Here some php code and results:

$cl = new SphinxClient ();
$cl->SetMatchMode (SPH_MATCH_EXTENDED2);
$cl->SetSortMode(SPH_SORT_RELEVANCE);
$cl->SetRankingMode(SPH_RANK_EXPR, 'sum(tf_idf*1000)');
$res = $cl->Query ( $tov_str, $index );
return $res;

so i'm, using "|" delimeters in my query and i noticed that when there's too much of arguments weight calculations sometimes (!) broke that caused wrong results. Just start a lot of querys and watch.
Look (pic1):
7 arguments and always 1 result
pic2 and pic3 :
add 1 argument and result could differ from each query.

0000464: @distinct value for setGroupDistinct on distributed index

Thu, 02 Feb 2012 04:14:21 +0100

I have a distributed index on 4 nodes. When i GroupBy a certain attribute and groupDistinct by another (just like the documentation http://www.sphinxsearch.com/docs/current.html#api-func-setgroupdistinct [^]), it returns the distinct count per index, and adds them together. This means the @distinct result is always 4 times the actual result.

The "quick" fix would be getting the max of the distinct counts, although this could be the wrong result as well.

0001080: seamless rotation is not really seamless

Wed, 01 Feb 2012 13:04:09 +0100

we have 2 machines: the master - mysql master, where we build sphinx indecies, the slave - mysql slave, where we just put the updated delta index from the master and call "kill -SIGHUP searchd_pid" to force searchd to rotate.
sometimes during the rotation on the slave searchd`s socket becomes broken with the following message:
"connection to /some/where/sphinx.sock failed ([Errno 61] Connection refused)"

for example, part of the searchd log (rotation is ok):

[Wed Feb 1 14:00:23.719 2012] [82790] rotating indices (seamless=1)
[Wed Feb 1 14:00:23.719 2012] [82790] DEBUG: CheckRotate invoked
[Wed Feb 1 14:00:23.720 2012] [82790] DEBUG: /some/where/sphinx/var/data/index.new.sph is not readable. Skipping
[Wed Feb 1 14:00:23.720 2012] [82790] DEBUG: Invoked SeamlessForkPrereader
[Wed Feb 1 14:00:23.720 2012] [82790] DEBUG: Invoked SeamlessTryToForkPrereader
[Wed Feb 1 14:00:23.720 2012] [82790] DEBUG: prealloc enough RAM and lock new index
[Wed Feb 1 14:00:23.723 2012] [82790] WARNING: rotating index: delta: '/some/where/sphinx/conf/wordforms.utf8.txt' differs from the original
[Wed Feb 1 14:00:23.723 2012] [82790] DEBUG: Locking the index via file /some/where/sphinx/var/data/index_delta.new.spl
[Wed Feb 1 14:00:23.723 2012] [82790] DEBUG: lock /some/where/sphinx/var/data/index_delta.new.spl success
[Wed Feb 1 14:00:23.723 2012] [82790] DEBUG: fork async reader
[Wed Feb 1 14:00:23.724 2012] [49395] DEBUG: CSphIndex_VLN::Preread invoked
[Wed Feb 1 14:00:23.724 2012] [49395] DEBUG: Prereading .spa
[Wed Feb 1 14:00:23.724 2012] [49395] DEBUG: Prereading .spm
[Wed Feb 1 14:00:23.733 2012] [49395] DEBUG: Prereading .sps
[Wed Feb 1 14:00:23.733 2012] [49395] DEBUG: Prereading .spk
[Wed Feb 1 14:00:23.734 2012] [49395] DEBUG: Prereading .spi
[Wed Feb 1 14:00:23.734 2012] [49395] DEBUG: Prereading .mvp
[Wed Feb 1 14:00:23.734 2012] [49395] DEBUG: Preread successfully finished
[Wed Feb 1 14:00:23.734 2012] [49395] DEBUG: SeamlessTryToForkPrereader: finishing the fork and invoking exit ( 0 )
[Wed Feb 1 14:00:24.667 2012] [82790] DEBUG: Base set to /some/where/sphinx/var/data/index_delta.old
[Wed Feb 1 14:00:24.773 2012] [82790] DEBUG: Base set to /some/where/sphinx/var/data/index_delta
[Wed Feb 1 14:00:25.068 2012] [82790] rotating index 'delta': success
[Wed Feb 1 14:00:25.071 2012] [82790] DEBUG: Unlocking the index (lock /some/where/sphinx/var/data/index_delta.old.spl)
[Wed Feb 1 14:00:25.071 2012] [82790] DEBUG: File ID ok, closing lock FD 10, unlinking /some/where/sphinx/var/data/index_delta.old.spl
[Wed Feb 1 14:00:25.154 2012] [82790] DEBUG: Invoked SeamlessForkPrereader
[Wed Feb 1 14:00:25.154 2012] [82790] rotating finished

at the very same time on the searcher side:
'82813 2012-02-01 14:00:23,791 DEBUG GET /SomeSearch/'
'82813 2012-02-01 14:00:23,791 INFO qstr: "admin4.dt00.net"'
'82810 2012-02-01 14:00:23,989 DEBUG GET /SomeSearch/'
'82810 2012-02-01 14:00:23,990 INFO qstr: "www.russianewsday.blogspot.com"'
'82819 2012-02-01 14:00:24,013 DEBUG GET /SomeSearch/'
'82819 2012-02-01 14:00:24,014 INFO qstr: "www.magazin-snov.ru"'
'82820 2012-02-01 14:00:24,199 DEBUG GET /SomeSearch/'
'82820 2012-02-01 14:00:24,200 INFO qstr: "www.blog.arbooz.com"'
'82812 2012-02-01 14:00:24,279 DEBUG GET /SomeSearch/'
'82812 2012-02-01 14:00:24,279 INFO qstr: "accounts.google.com"'
'82817 2012-02-01 14:00:24,288 DEBUG GET /SomeSearch/'
'82817 2012-02-01 14:00:24,288 INFO qstr: "www.passport.bigmir.net"'
'82814 2012-02-01 14:00:24,343 DEBUG GET /SomeSearch/'
'82814 2012-02-01 14:00:24,343 INFO qstr: "v-mire.com"'
'82806 2012-02-01 14:00:24,422 DEBUG GET /SomeSearch/'
'82806 2012-02-01 14:00:24,423 INFO qstr: "house-sk.ru"'
'82809 2012-02-01 14:00:24,448 DEBUG GET /SomeSearch/'
'82809 2012-02-01 14:00:24,448 INFO qstr: "linkfromblog.com"'
'82818 2012-02-01 14:00:24,624 DEBUG GET /SomeSearch/'
'82818 2012-02-01 14:00:24,625 INFO qstr: "www.isras.ru"'
'82808 2012-02-01 14:00:24,692 DEBUG GET /SomeSearch/'
'82808 2012-02-01 14:00:24,692 INFO qstr: "www.moscowsg.megafon.ru"'
'82807 2012-02-01 14:00:24,719 DEBUG GET /SomeSearch/'
'82807 2012-02-01 14:00:24,720 INFO qstr: "seks.am"'
'82809 2012-02-01 14:00:24,752 ERROR Searcher::querySphinx error: in RunQueries connection to daemon via socket failed: connection to /some/where/log/sphinx.sock failed ([Errno 61] Connection refused), warning: '
'82809 2012-02-01 14:00:24,752 INFO build matches in 0.303702 sec'
'82809 2012-02-01 14:00:24,752 INFO build snippets in 3.09944e-06 sec'
'82809 2012-02-01 14:00:24,753 INFO query total in 0.303705 sec (0.304371) found 0'
'82818 2012-02-01 14:00:24,929 ERROR Searcher::querySphinx error: in RunQueries connection to daemon via socket failed: connection to /some/where/log/sphinx.sock failed ([Errno 61] Connection refused), warning: '
'82818 2012-02-01 14:00:24,929 INFO build matches in 0.304593 sec'
'82818 2012-02-01 14:00:24,930 INFO build snippets in 4.05312e-06 sec'
'82818 2012-02-01 14:00:24,930 INFO query total in 0.304597 sec (0.305263) found 0'
'82808 2012-02-01 14:00:24,996 ERROR Searcher::querySphinx error: in RunQueries connection to daemon via socket failed: connection to some/where/log/sphinx.sock failed ([Errno 61] Connection refused), warning: '
'82808 2012-02-01 14:00:24,996 INFO build matches in 0.303465 sec'
'82808 2012-02-01 14:00:24,996 INFO build snippets in 3.09944e-06 sec'
'82808 2012-02-01 14:00:24,997 INFO query total in 0.303468 sec (0.304168) found 0'
'82807 2012-02-01 14:00:25,024 ERROR Searcher::querySphinx error: in RunQueries connection to daemon via socket failed: connection to some/where/log/sphinx.sock failed ([Errno 61] Connection refused), warning: '
'82807 2012-02-01 14:00:25,024 INFO build matches in 0.303946 sec'
'82807 2012-02-01 14:00:25,024 INFO build snippets in 4.05312e-06 sec'
'82807 2012-02-01 14:00:25,025 INFO query total in 0.30395 sec (0.304612) found 0'
'82818 2012-02-01 14:00:25,060 DEBUG GET /SomeSearch/'
'82818 2012-02-01 14:00:25,061 INFO qstr: "www.bitshare.com"'
'82807 2012-02-01 14:00:25,092 DEBUG GET /SomeSearch/'
'82807 2012-02-01 14:00:25,093 INFO qstr: "www.passport.bigmir.net"'
'82813 2012-02-01 14:00:25,160 INFO build matches in 1.36811 sec'
'82813 2012-02-01 14:00:25,160 INFO build snippets in 7.86781e-06 sec'
'82813 2012-02-01 14:00:25,160 INFO query total in 1.36812 sec (1.36893) found 0'
'82819 2012-02-01 14:00:25,161 INFO build matches in 1.14664 sec'
'82819 2012-02-01 14:00:25,161 INFO build snippets in 8.10623e-06 sec'
'82819 2012-02-01 14:00:25,161 INFO query total in 1.14665 sec (1.14756) found 0'

exectly during the rotation some searcher forks are unable to connect to sphinx socket.

part of searchd config:
searchd
{
listen = /some/where/log/sphinx.sock
...
seamless_rotate = 1
preopen_indexes = 1
unlink_old = 1
workers = fork
}

it seems that during seamless rotation in some very short amount of time searchd doesn`t listen the socket, and the searcher gets "connection refused".

is it possible to fix that somehow?

we use python api.

0001079: random RT related (?) crashes on ubuntu 11.10 with gcc 4.6

Wed, 01 Feb 2012 01:00:55 +0100

A test Python program doing R/W workload is reported to cause random searchd crashes as of r3092, branches/rel20. Those only happen on Ubuntu 11.10 x64 (kernel 3.0.x, gcc 4.6.x) and won't repeat on Ubuntu 10.04 LTS or Windows on my testing. The test data is on FTP.

0001075: Crash on using RT indexes on 2.0.3-release

Mon, 30 Jan 2012 13:53:08 +0100

------- FATAL: CRASH DUMP -------
[Sun Jan 29 02:04:45.635 2012] [15060]

--- crashed SphinxAPI request dump ---
AAEBAwAAAHEAAAAAAAAAAQAAABZwcm9kX2JlenBsYXRub19hZHZlcnRzAAAACjA4NzY2MzkxNjkAAAADPGI+AAAA
BDwvYj4AAAAFIC4uLiAAAAEAAAAABQAAAAAAAAAAAAAAAQAAAAVpbmRleAAAAARub25lAAAAAA==
--- request dump end ---
Sphinx 2.0.3-release (r3043)
-------------- backtrace begins here ---------------
Program compiled with gcc 4.4.3
Host OS is Linux icepique 2.6.32-37-server 0000081-Ubuntu SMP Fri Dec 2 20:49:12 UTC 2011 x86_64 GNU/Linux
Stack bottom = 0x7f70b2908e7f, thread stack size = 0x14000
begin of manual backtrace:
Something wrong with thread stack, backtrace may be incorrect (fp=7f70b2905360)
Stack looks OK, attempting backtrace.
413cc3
44a678
44c569
4547d2
4549b9
54660d
7f70c5e829ca
Something wrong in frame pointers, backtrace failed (fp=0)
begin of system backtrace:
begin of system symbols:
/opt/sphinx/bin/searchd[0x53de24]
/opt/sphinx/bin/searchd(_ZN16SphCrashLogger_c11HandleCrashEi+0x193)[0x413cc3]
/lib/libpthread.so.0(+0xf8f0)[0x7f70c5e8b8f0]
/opt/sphinx/bin/searchd[0x44a678]
/opt/sphinx/bin/searchd[0x44c569]
/opt/sphinx/bin/searchd[0x4547d2]
/opt/sphinx/bin/searchd(_Z13HandlerThreadPv+0x59)[0x4549b9]
/opt/sphinx/bin/searchd(_Z20sphThreadProcWrapperPv+0x1d)[0x54660d]
/lib/libpthread.so.0(+0x69ca)[0x7f70c5e829ca]
/lib/libc.so.6(clone+0x6d)[0x7f70c4be570d]
-------------- backtrace ends here ---------------
--- 2 active threads ---
thd 0, proto sphinxql, state net_read, command delete
thd 1, proto sphinxapi, state query, command excerpt
------- CRASH DUMP END -------

0001061: Cannot allocate memory (length=17179869180 is over 2GB)

Mon, 30 Jan 2012 03:03:15 +0100

I don't use RT indexes.

[Wed Jan 18 23:39:43.454 2012] [20819] FATAL: sphFDSet() failed fd=1031, FD_SETSIZE=1024
[Wed Jan 18 23:39:43.458 2012] [20819] FATAL: sphFDSet() failed fd=1066, FD_SETSIZE=1024
[Wed Jan 18 23:39:43.463 2012] [20819] WARNING: failed to send server version (client=10.41.47.XXX:46666)
[Wed Jan 18 23:39:43.469 2012] [20819] FATAL: sphFDSet() failed fd=1076, FD_SETSIZE=1024
[Wed Jan 18 23:39:43.478 2012] [20819] FATAL: sphFDSet() failed fd=1081, FD_SETSIZE=1024
[Wed Jan 18 23:39:43.488 2012] [20819] FATAL: sphFDSet() failed fd=1086, FD_SETSIZE=1024
[Wed Jan 18 23:39:43.500 2012] [20819] FATAL: sphFDSet() failed fd=1091, FD_SETSIZE=1024
[Wed Jan 18 23:39:43.503 2012] [20819] FATAL: sphFDSet() failed fd=1096, FD_SETSIZE=1024
[Wed Jan 18 23:39:43.506 2012] [20819] FATAL: sphFDSet() failed fd=1101, FD_SETSIZE=1024
[Wed Jan 18 23:39:43.519 2012] [20819] FATAL: sphFDSet() failed fd=1106, FD_SETSIZE=1024
[Wed Jan 18 23:39:43.544 2012] [20819] FATAL: sphFDSet() failed fd=1111, FD_SETSIZE=1024
[Wed Jan 18 23:39:43.548 2012] [20819] FATAL: sphFDSet() failed fd=1117, FD_SETSIZE=1024
[Wed Jan 18 23:39:43.557 2012] [20819] FATAL: sphFDSet() failed fd=1125, FD_SETSIZE=1024
[Wed Jan 18 23:39:44.812 2012] [20819] WARNING: send() failed: 9: Bad file descriptor
[Wed Jan 18 23:39:44.844 2012] [20819] WARNING: failed to send server version (client=10.41.47.XXX:50597)
[Wed Jan 18 23:39:44.845 2012] [20819] WARNING: send() failed: 9: Bad file descriptor
[Wed Jan 18 23:39:44.903 2012] [20819] WARNING: send() failed: 32: Broken pipe
[Wed Jan 18 23:39:45.346 2012] [20819] WARNING: send() failed: 9: Bad file descriptor
[Wed Jan 18 23:39:46.483 2012] [20819] WARNING: last message repeated 1 times
[Wed Jan 18 23:39:46.483 2012] [20819] WARNING: thread missing from thread table
[Wed Jan 18 23:39:47.448 2012] [20819] WARNING: last message repeated 3 times
[Wed Jan 18 23:39:47.448 2012] [20819] WARNING: send() failed: 9: Bad file descriptor
[Wed Jan 18 23:39:47.501 2012] [20819] WARNING: thread missing from thread table
[Wed Jan 18 23:39:48.531 2012] [20819] WARNING: last message repeated 18 times
[Wed Jan 18 23:39:48.531 2012] [20819] WARNING: index alphathread: attrs save failed: rollback rename to '/var/lib/sphinx/forum/alphathread.spa' failed: No such file or directory; INDEX UNUSABLE; FIX FILE NAMES MANUALLY
[Wed Jan 18 23:39:48.532 2012] [20819] WARNING: last message repeated 1 times
[Wed Jan 18 23:39:48.532 2012] [20819] WARNING: thread missing from thread table
....
-------------- backtrace ends here ---------------
--- 5 active threads ---
thd 0, proto sphinxql, state net_read, command insert
thd 1, proto sphinxql, state net_read, command insert
thd 2, proto sphinxql, state net_read, command insert
thd 3, proto sphinxapi, state query, command update
thd 4, proto sphinxapi, state query, command update
------- CRASH DUMP END -------
[Thu Jan 19 00:08:32.733 2012] [21514] Child process 28014 has been finished by CRASH_EXIT (exit code 2), will be restarted
[Thu Jan 19 00:08:32.733 2012] [21514] Child process 1301 has been forked
[Thu Jan 19 00:08:32.737 2012] [ 1301] WARNING: index 'alphathread': preload: mmap() failed: Cannot allocate memory (length=17179869180 is over 2GB, impossible on some
32-bit systems); NOT SERVING
[Thu Jan 19 00:08:34.336 2012] [ 1301] accepting connections
------- FATAL: CRASH DUMP -------
[Thu Jan 19 00:08:36.733 2012] [ 1301]

--- crashed SphinxAPI request dump ---
AAIBAgAAAGUAAAAWYWxwaGF0aHJlYWQsc2NhbnRocmVhZAAAAAMAAAAIbGFzdHBvc3QAAAAAAAAACnJlcGx5Y291
bnQAAAAAAAAABXZpZXdzAAAAAAAAAAEAAAAAABYr508Xs2QAAAADAAAAJg==
--- request dump end ---
Sphinx 2.0.2-id64-dev (rel20-r3030)
-------------- backtrace begins here ---------------
Program compiled with gcc 4.4.4
Host OS is Linux apollo.axivo.com 2.6.32-71.29.1.el6.x86_64 0000001 SMP Mon Jun 27 19:49:27 BST 2011 x86_64 x86_64 x86_64 GNU/Linux
Stack bottom = 0x7f627751bdbf, thread stack size = 0x44000
begin of manual backtrace:
Something wrong with thread stack, backtrace may be incorrect (fp=7f627751bdbf)
Stack looks OK, attempting backtrace.
0
Something wrong in frame pointers, backtrace failed (fp=0)
begin of system backtrace:
begin of system symbols:
/lib64/libpthread.so.0[0x3ff300f4a0]
/usr/bin/searchd(_ZNK13CSphIndex_VLN11FindDocinfoEm+0x3d)[0x4be3a5]
/usr/bin/searchd(_ZN13CSphIndex_VLN16UpdateAttributesERK14CSphAttrUpdateiR10CSphString+0x1821)[0x4db2a3]
/usr/bin/searchd[0x47ba9a]
/usr/bin/searchd(_Z19HandleCommandUpdateiiR13InputBuffer_c+0x16e3)[0x4a3720]
/usr/bin/searchd(_Z18HandleClientSphinxiPKcP9ThdDesc_t+0x45a)[0x4a7efd]
/usr/bin/searchd(_Z13HandlerThreadPv+0x57)[0x4a8731]
/usr/bin/searchd(_Z20sphThreadProcWrapperPv+0x19)[0x584195]
/lib64/libpthread.so.0[0x3ff30077f1]
/lib64/libc.so.6(clone+0x6d)[0x3ff2ce570d]
-------------- backtrace ends here ---------------
--- 4 active threads ---
thd 0, proto sphinxapi, state query, command update
thd 1, proto sphinxql, state net_read, command insert
thd 2, proto sphinxql, state net_read, command insert
thd 3, proto sphinxapi, state query, command update
------- CRASH DUMP END -------
[Thu Jan 19 00:08:36.963 2012] [21514] Child process 1301 has been finished by CRASH_EXIT (exit code 2), will be restarted
[Thu Jan 19 00:08:36.963 2012] [21514] Child process 2310 has been forked
[Thu Jan 19 00:08:36.966 2012] [ 2310] WARNING: index 'alphathread': preload: mmap() failed: Cannot allocate memory (length=17179869180 is over 2GB, impossible on some 32-bit systems); NOT SERVING
[Thu Jan 19 00:08:38.571 2012] [ 2310] accepting connections
[Thu Jan 19 00:10:05.710 2012] [ 2310] caught SIGTERM, shutting down
[Thu Jan 19 00:10:06.063 2012] [ 2310] shutdown complete
[Thu Jan 19 00:12:29.133 2012] [ 3576] Child process 3577 has been forked
[Thu Jan 19 00:12:29.136 2012] [ 3577] WARNING: index 'alphathread': preload: mmap() failed: Cannot allocate memory (length=17179869180 is over 2GB, impossible on some 32-bit systems); NOT SERVING
[Thu Jan 19 00:12:30.715 2012] [ 3577] accepting connections
[Thu Jan 19 00:13:22.154 2012] [ 3577] caught SIGTERM, shutting down
[Thu Jan 19 00:13:22.481 2012] [ 3577] shutdown complete
[Thu Jan 19 00:13:23.420 2012] [ 3857] Child process 3858 has been forked
[Thu Jan 19 00:13:23.423 2012] [ 3858] WARNING: index 'alphathread': preload: mmap() failed: Cannot allocate memory (length=17179869180 is over 2GB, impossible on some 32-bit systems); NOT SERVING
[Thu Jan 19 00:13:25.083 2012] [ 3858] accepting connections

Kernel Logs:
Jan 19 11:38:35 leo kernel: searchd[31701]: segfault at 0 ip 000000000048b625 sp 00007f062609dbb0 error 4
Jan 19 11:38:35 leo kernel: searchd[31742]: segfault at 0 ip 000000000048b625 sp 00007f06255b9bb0 error 4
Jan 19 11:38:35 leo kernel: searchd[31657]: segfault at 7f06446e3700 ip 00007f06446e3700 sp 00007f0626c4dba8 error 15
Jan 19 11:38:35 leo kernel: searchd[1113]: segfault at 7f06ff1fb000 ip 00000000004e0bf9 sp 00007f0625f524b0 error 4
Jan 19 11:38:35 leo kernel: in searchd[400000+27d000]
Jan 19 11:38:35 leo kernel: in searchd[400000+27d000]
Jan 19 11:38:35 leo kernel: searchd[1144]: segfault at 3200000000 ip 0000003ff2d24716 sp 00007f065edd6178 error 4
Jan 19 11:38:35 leo kernel: searchd[1146]: segfault at 7f06ff1fb000 ip 00000000004e0bf9 sp 00007f0624bee4b0 error 4
Jan 19 11:38:35 leo kernel: searchd[1135]: segfault at 7f06ff1fb000 ip 00000000004e0bf9 sp 00007f062601e4b0 error 4 in libc-2.12.so[3ff2c00000+197000] in searchd[400000+27d000] in searchd[400000+27d000]
Jan 19 11:38:35 leo kernel:
Jan 19 11:38:35 leo kernel:
Jan 19 11:38:35 leo kernel: searchd[1136]: segfault at 7f06ff1fb000 ip 00000000004e0bf9 sp 00007f0644bee4b0 error 4 in searchd[400000+27d000]
Jan 19 11:38:35 leo kernel: in searchd[400000+27d000]

0001066: BuildExcerpts failure with large dataset and/or many HTML tags

Mon, 30 Jan 2012 01:24:27 +0100

On Sphinx 2.0.3, when trying to build excerpts with the BuildExcerpts() call (via the Python API), searchd shows error:

WARNING: failed to receive client request body

and BuildExcerpts() will raise exception:

error: unpack requires a string argument of length 8

with the error occurring on this line:

(status, ver, length) = unpack('>2HL', sock.recv(8))

After some testing, it looked like this only occurred when the source documents were large in nature or contained very large amounts of HTML tags.

0001072: WARNING: send() failed: 32: Broken pipe

Sat, 28 Jan 2012 21:20:28 +0100

While using r3030 in CentOS 6.2 64 bits, I get the following errors related to broken TCP pipes:

[Thu Jan 26 03:15:03.508 2012] [13672] log reopened
[Thu Jan 26 03:15:03.508 2012] [13672] query log reopened
[Thu Jan 26 03:16:16.338 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 03:20:03.350 2012] [13672] WARNING: send() failed: 32: Broken pipe
[Thu Jan 26 03:20:04.779 2012] [13672] WARNING: last message repeated 2 times
[Thu Jan 26 03:24:16.558 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 03:30:53.738 2012] [13672] WARNING: send() failed: 32: Broken pipe
[Thu Jan 26 03:30:55.154 2012] [13672] WARNING: last message repeated 2 times
[Thu Jan 26 03:32:16.727 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 03:32:18.608 2012] [13672] WARNING: send() failed: 32: Broken pipe
[Thu Jan 26 03:32:20.165 2012] [13672] WARNING: last message repeated 4 times
[Thu Jan 26 03:40:16.674 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 03:45:32.218 2012] [13672] WARNING: send() failed: 32: Broken pipe
[Thu Jan 26 03:48:16.702 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 03:56:17.117 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 04:00:16.573 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 04:08:17.356 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 04:08:19.630 2012] [13672] WARNING: send() failed: 32: Broken pipe
[Thu Jan 26 04:08:20.891 2012] [13672] WARNING: last message repeated 1 times
[Thu Jan 26 04:15:31.052 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 04:16:02.446 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 04:16:25.189 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 04:16:42.355 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 04:16:49.640 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 04:16:56.338 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 04:17:02.518 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 04:24:02.165 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 04:32:02.453 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 04:40:02.209 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 04:48:02.037 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 04:56:02.028 2012] [13672] rotating indices (seamless=1)
[Thu Jan 26 04:56:48.426 2012] [13672] WARNING: send() failed: 32: Broken pipe
[Thu Jan 26 04:56:49.755 2012] [13672] WARNING: last message repeated 1 times
[Thu Jan 26 05:00:02.026 2012] [13672] rotating indices (seamless=1)

0001074: Can't compile sphinxsearch against percona 5.5

Sat, 28 Jan 2012 16:10:22 +0100

When trying to compile sphinxsearch-2.0.3 (but same thing with 2.0.2 and according to google groups, also to 0.9.9 (http://groups.google.com/group/percona-discussion/browse_frm/thread/114ecbed8c77c23e/2a85d8c4010aaf57?lnk=gst&q=sphinxstd#2a85d8c4010aaf57 [^]) ) with percona server, we got the following error :

/usr/lib/libmysqlclient.a(my_new.cc.o): In function `operator delete[](void*)':
/home/jenkins/workspace/percona-server-5.5-debs/label_exp/debian6-64/target/Percona-Server-5.5.18-rel23.0/mysys/my_new.cc:43: multiple definition of `operator delete[](void*)'
libsphinx.a(sphinxstd.o):/root/boursorama-sphinxsearch/build/boursorama-sphinxsearch/src/sphinxstd.cpp:615: first defined here
/usr/lib/libmysqlclient.a(my_new.cc.o): In function `operator delete(void*)':
/home/jenkins/workspace/percona-server-5.5-debs/label_exp/debian6-64/target/Percona-Server-5.5.18-rel23.0/mysys/my_new.cc:37: multiple definition of `operator delete(void*)'
libsphinx.a(sphinxstd.o):/root/boursorama-sphinxsearch/build/boursorama-sphinxsearch/src/sphinxstd.cpp:608: first defined here
/usr/lib/libmysqlclient.a(my_new.cc.o): In function `operator new[](unsigned long)':
/home/jenkins/workspace/percona-server-5.5-debs/label_exp/debian6-64/target/Percona-Server-5.5.18-rel23.0/mysys/my_new.cc:32: multiple definition of `operator new[](unsigned long)'
libsphinx.a(sphinxstd.o):/root/boursorama-sphinxsearch/build/boursorama-sphinxsearch/src/sphinxstd.cpp:600: first defined here
/usr/lib/libmysqlclient.a(my_new.cc.o): In function `operator new(unsigned long)':
/home/jenkins/workspace/percona-server-5.5-debs/label_exp/debian6-64/target/Percona-Server-5.5.18-rel23.0/mysys/my_new.cc:27: multiple definition of `operator new(unsigned long)'
libsphinx.a(sphinxstd.o):/root/boursorama-sphinxsearch/build/boursorama-sphinxsearch/src/sphinxstd.cpp:591: first defined here

0001073: Не срабатывают стоп-слова если они есть в словоформах

Sat, 28 Jan 2012 09:27:45 +0100

Есть стоп-слова : ЗАО, ООО, предлоги и прочие
Есть словоформы из myspell

Если стоп-слово есть в словоформах оно попадает в индекс и по нему идёт поиск

Sphinx 2.0.3-id64-release (r3043), собирался и без --enable-id64 результат одинаковый

0001058: Corrupt results with Distributed RT index queries via SphinxAPI

Fri, 27 Jan 2012 10:51:43 +0100

This is a somewhat complicated. We've been testing sphinx quite successfully, and today switched from a single real time index to 4 real time distributed indexes (local machine only).

Using sphinx 2.0.3 from source.

Using the SphinxAPI, we're receiving some, what appears to be, corrupt results: http://pastebin.com/S7NnWAUQ [^] Line 382 and 383.

I tend to think this is a bug, as the same setup was working before I change my configuration over to distributed queries. I wanted to bring this up in the forum before I lodge a bug report.

0001069: [PHP API] Incorrect unpacking of document id's when using the BCMath extension

Tue, 24 Jan 2012 15:41:53 +0100

The sphinxapi.php sphUnpack64 function is dependent on the global bcscale setting. If this settings is larger than 0, unpacked 64 bit document id's are returned as numbers with decimal points even though they should be integer values.

0001064: YEARMONTHDAY is off by about 47 seconds in 2012

Fri, 20 Jan 2012 17:58:03 +0100

The YEARMONTHDAY function translates a timestamp into a Year / Month / Day value using localtime_r().

The output of YEARMONTHDAY is 47 seconds ahead of what a Linux/POSIX system call would report.

This is really a big annoyance, because when you group search hits by day, some hits are misclassified and when users count the real hits they notice a difference.

0001011: MVA negative values converted to positive

Fri, 20 Jan 2012 15:12:37 +0100

When importing source trough xmlpip2 with schema supplied in it - there is no way to get negative values in multi-valued attributes.

0001062: Incorrect search at bigint fields using SphinxQL

Fri, 20 Jan 2012 11:32:55 +0100

I use sql_attr_bigint or sql_attr_multi(bigint) in my index. I try to do search in that
fields. All OK then I find values less then 2^32. But if one of limit in BETWEEN is more
then 2^32 - results doesn't contain big values. In that case query in querys-log is
different my.

0001045: sphinx UDF error

Tue, 17 Jan 2012 17:02:30 +0100

I hve to use udf in my application,
I tried it in the sample database(test.sql) provided along with the sphinx tar file.
The create function works fine but when i call the function in query it throws error as

ERROR 2013 (HY000): Lost connection to MySQL server during query

I have copied the content of my example file in additional description.

0001051: UDF does not support MVA BIGINT

Mon, 16 Jan 2012 20:08:42 +0100

When we try to send MVA BIGINT as parameter to User Defined Function (UDF) we get the following error:
internal error: unmapped UDF argument type (arg=0, type=1073741826).

MVA UINT works ok.

0001056: RT index searching order of disk vs ram chunk

Mon, 16 Jan 2012 19:39:06 +0100

searching of RT index begins with oldest disk chunk then ram chunk.

In case max_query_time or cutoff limit got hit we might get inaccurate results.

It should be better to reorganize code to search ram chunks first then disk chunk but from newest to oldest. It possible as with proper setup and usage of kill-list that already got implemented.

0001052: Range Filtering Number Attributes Starting with Zero

Mon, 16 Jan 2012 17:30:14 +0100

(Essentially copied from http://sphinxsearch.com/forum/view.html?id=8890 [^] )

(This is actually a problem in 2.0.3, too, but it wasn't an option in the version dropdown).

I was doing a search just recently (via SphinxSE if it matters) where I was filtering number attributes by range. These numbers represent ZIP codes and so these numbers can start with zeroes.

When passing a number beginning a zero as an attribute range filter to Sphinx, Sphinx returns a different result set than if filtering without the prefixed zero, but only if part of the filtered range 'value' falls outside what appears to be valid octal numbers. For example, filtering a range such as '01234-01235' works, but using '08765-08764' fails.

If I do the search as '8765-8764' it returns the values expected and using a normal (non-range) filter of say '08765' works as expected, too.

0001041: Searchd 2.0.3 doesn't seems to be able to work with prefork

Thu, 12 Jan 2012 12:44:53 +0100

Hi, I recently enabled prefork on sphinx 2.0.3 release having previously
tested it (worked ok) on 1.1 beta.

It is enabled using 'worker = prefork' in the sphinx.conf file.

When I start sphinx I observe fork like behavior, it starts only the main
process and forks new ones on demand and they shutdown again when not
needed, when I previously tested on 1.1 beta it started numerous processes
and they stayed running all the time.

Is there a new undocumented configuration related to this? or is it possible
I have come across a bug.

0001040: Searchd doesn't start with threads

Thu, 12 Jan 2012 12:41:16 +0100

Always when I try to start with workers=threads I am getting

root@server4 ~ # sudo -u _sphinx /usr/local/sphinx-svn/sphinx/bin/searchd --config /usr/local/sphinx-svn/sphinx/etc/sphinx.conf --iostats --cpustats
Sphinx 2.0.3-release (r3043)
Copyright (c) 2001-2011, Andrew Aksyonoff
Copyright (c) 2008-2011, Sphinx Technologies Inc (http://sphinxsearch.com [^])

using config file '/usr/local/sphinx-svn/sphinx/etc/sphinx.conf'...
listening on all interfaces, port=9313
root@server4 ~ # listening on 127.0.0.1:9307
WARNING: process shared mutex unsupported, MVA update disabled ( pthread_mutexattr_setpshared, errno = 22 )
precaching index 'i1'
precaching index 'i2'
rotating index 'i2': success
precached 2 indexes in 0.026 sec

And searchd stays running without to return to command line. Any other workers model works...

0001009: Delay when PHP connector (mysqli, mysql) closes a connexion

Wed, 11 Jan 2012 14:13:53 +0100

When I close an SQL connexion in PHP, the socket appears UP.

I think the socket is really closed after a timeout.