This sounds really strange.

We know that Jackrabbit has performance impacts on heavy write load. 
Can you provide more information's about how many save operations occur ..

The GLOBAL_REVISION Table is used a lock in a clustered environment
so its ok to have locks there but so long ...

greets
claus

-----Ursprüngliche Nachricht-----
Von: Davy Kamerbeek [mailto:i...@kamerbeekict.nl] 
Gesendet: Dienstag, 20. September 2016 11:14
An: dev@jackrabbit.apache.org
Betreff: JCR session.save() bad performance in clustered environment

Hi All,

We are currently using database journal in the jackrabbit cluster (6 JBoss4
nodes with NFS storage). We have a high amount document upload through all
the nodes, which includes concurrent users. We are facing long blocks on
J_C_GLOBAL_REVISION table. These blocks were typically ranging from 10 to 45
minutes duration and documents could not be uploaded in this duration. Short
blocks of few seconds were also very frequently observed:

>>      19273 | jcr_journaal | update J_C_GLOBAL_REVISION set REVISION_ID =
>> REVISION_ID + 1 | 00:00:02.143491   |        22873 | jcr_journaal  |
>> update J_C >> _GLOBAL_REVISION set REVISION_ID = REVISION_ID + 1 |
>> 00:00:22.469581
>>      22873 | jcr_journaal | update J_C_GLOBAL_REVISION set REVISION_ID =
>> REVISION_ID + 1 | 00:00:22.469581   |        19273 | jcr_journaal  |
>> update J_C >> _GLOBAL_REVISION set REVISION_ID = REVISION_ID + 1 |
>> 00:00:02.143491
>>      19273 | jcr_journaal | <IDLE> in transaction                            
>>            
>> | 00:00:02.141283   |        28491 | jcr_journaal  | update J_C >>
>> _GLOBAL_REVISION set REVISION_ID = REVISION_ID + 1 | 00:00:22.469581
>>      22873 | jcr_journaal | <IDLE> in transaction                            
>>            
>> | 00:00:02.141283   |        28491 | jcr_journaal  | update J_C >>
>> _GLOBAL_REVISION set REVISION_ID = REVISION_ID + 1 | 00:00:02.143491

<?xml version="1.0"?>
<!DOCTYPE Repository PUBLIC &quot;-//The Apache Software Foundation//DTD
Jackrabbit 1.5//EN&quot;
                           
&quot;http://jackrabbit.apache.org/dtd/repository-1.5.dtd&quot;>



The average storing time (JCR Session -->session.save()) is 5000ms -
20000ms. But with peak load we are getting blocks and all nodes are waiting
to get through. Somehow clustering is not working properly.

I appreciate any help!




--
View this message in context: 
http://jackrabbit.510166.n4.nabble.com/JCR-session-save-bad-performance-in-clustered-environment-tp4664815.html
Sent from the Jackrabbit - Dev mailing list archive at Nabble.com.

Reply via email to