Dovecot+Solr+Tika segmentation fault

[John Fawcett]

On 04/12/2020 09:29, Manuel Josupeit-Walter wrote:
>>>> A fix was posted on this list which seems to avoid segfaults when
>>>> using solr+tika. I reposted it not long ago in response to another
>>>> question. Before exploring the other segfaults it would be worthwhile
>>>> getting that fix. If other segfaults continue then I suggest posting
>>>> the logging from Dovecot
>>> https://github.com/dovecot/core/commit/043d111cea78533cf22d3bfe0872c8376d842f56[https://github.com/dovecot/core/commit/043d111cea78533cf22d3bfe0872c8376d842f56]
>>> (DOV-4090)
>>>
>>> Scheduled for 2.3.13.
>> I applied the fix and will observe it for the next couple of days.
> Unfortunately, the issue still persists even with this patch. The log
> message is slightly different, though, regarding disconnection:
>
> Dec 03 17:10:49 server.name tika[988]: INFO  tika/ (application/pdf)
> Dec 03 17:10:49 server.name tika[988]: INFO  tika/ (image/png)
> Dec 03 17:10:49 server.name kernel: indexer-worker[15805]: segfault at 0 ip 00007f140483db64 sp 00007ffc04fa7960 error 4 in libdovecot.so.0.0.0[7f14047f3000+db000]
> Dec 03 17:10:49 server.name kernel: Code: 89 f5 48 8d b8 f0 01 00 00 48 8b 52 08 4c 8b 62 48 e8 90 89 fb ff 48 8b 44 24 08 be 00 20 00 00 4c 89 ef 48 8b 80 10 01 00 00 <4c> 8b 38 e8 64 b2 fb ff 48 8b 44 24 0
> Dec 03 17:10:49 server.name dovecot[988]: Dec 03 16:10:49 indexer: Error: Indexer worker disconnected, discarding 1 requests for username_stripped
> Dec 03 17:10:49 server.name dovecot[988]: Dec 03 16:10:49 indexer-worker(username_stripped)<1952><Ed3ek5G1BMQgARa4EL8DAKGMINRlZjjS:c+gACAkOyV+gBwAAhFs1yA>: Fatal: master: service(indexer-worker): child 1952 killed
>
> It seems that there is something else going wrong in this version.
>
> Best,
> Manuel

Also just to clarify the above mentioned patch solves an assert which
happens when using solr.

There is another patch which solves an assert when using solr+tika
together: that one is not mine but I'll cite it here to make it easier
to identify it. When running solr+tika together I found I needed both
patches. I first saw this in an email to this list from Josef 'Jeff'
Sipek on 19/08/2020.

diff --git a/src/plugins/fts-solr/solr-connection.c
b/src/plugins/fts-solr/solr-connection.c
index
ae720b5e2870a852c1b6c440939e3c7c0fa72b5c..9d364f93e2cd1b716b9ab61bd39656a6c5b1ea04
100644
--- a/src/plugins/fts-solr/solr-connection.c
+++ b/src/plugins/fts-solr/solr-connection.c
@@ -103,7 +103,7 @@ int solr_connection_init(const struct
fts_solr_settings *solr_set,
http_set.ssl = ssl_client_set;
http_set.debug = solr_set->debug;
http_set.rawlog_dir = solr_set->rawlog_dir;
- solr_http_client = http_client_init(&http_set);
+ solr_http_client = http_client_init_private(&http_set);
}
*conn_r = conn;
diff --git a/src/plugins/fts/fts-parser-tika.c
b/src/plugins/fts/fts-parser-tika.c
index
a4b8b5c3034f57e22e77caa759c090da6b62f8ba..b8b57a350b9a710d101ac7ccbcc14560d415d905
100644
--- a/src/plugins/fts/fts-parser-tika.c
+++ b/src/plugins/fts/fts-parser-tika.c
@@ -77,7 +77,7 @@ tika_get_http_client_url(struct mail_user *user,
struct http_url **http_url_r)
http_set.request_timeout_msecs = 60*1000;
http_set.ssl = &ssl_set;
http_set.debug = user->mail_debug;
- tika_http_client = http_client_init(&http_set);
+ tika_http_client = http_client_init_private(&http_set);
}
*http_url_r = tuser->http_url;
return 0;

John