You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@jackrabbit.apache.org by Ian Boston <ie...@tfd.co.uk> on 2010/03/05 12:08:39 UTC

UserManager peformance in 1.6

Hi,
I am trying to optimise a bit of code that adds users to jackrabbit, I have a lot of users to load.

Are these times unreasonable for Jackrabbit.

It takes 15ms to get the user with a call like 

      String principalName = "newUserId"
      Authorizable authorizable = userManager.getAuthorizable(principalName);

user does not exist.

It takes 44ms to create a user with a call like

        User user = userManager.createUser(principalName, pass, p, prefix);

prefix is something line "aa/bb/cc/dd"

aa,bb,cc,dd all exist, and should be in memory and dd has < 255 children

I also create about 5 other nodes add some simple ACL's and the whole operation takes upto 250ms.

I am a bit worried, since on in other RDBMS based  systems, that this will be compared against, this sort of operation is faster,

deployers need to add in the range 10K to 4M, at the top end thats going to take11 days to load if there is no slow down.

Any suggestions how to make node creation faster?
Is there some magic trick that I am missing?

Jackrabbit 1.6
Java 1.6
OOTB standard config with Derby as a back end.
MacBook Pro, 2.53 Ghz Intel Core 2 Duo 4GB DDR3


Thanks
Ian

Re: UserManager peformance in 1.6

Posted by Ian Boston <ie...@tfd.co.uk>.
I guess this is related to  https://issues.apache.org/jira/browse/JCR-2199 
If it is the problem goes away in JR2
asking on dev.
Ian

On 5 Mar 2010, at 12:19, Ian Boston wrote:

> After doing some profiling, I think the reason is that where a user does not exist search has to be used to verify that it does not exist prior to adding the user. Using Yourkit I see about 55% of the time spent in the Lucene index performing a query triggered in UserManagerImpl.getUserNode(userId).
> 
> I assume that this query is needed because user nodes can exist in many places in in the repo under /rep:security/rep:authorizables so the path cannot be inferred from the username ?
> 
> Was there some design criteria that made this necessary, and is there any way to avoid the search on create.
> 
> I think this search must happen in every session that hasn't seen the user before which must be slowing almost everything where these isnt a pooled session down ?
> 
> Ian
> 


Re: UserManager peformance in 1.6

Posted by Ian Boston <ie...@tfd.co.uk>.
After doing some profiling, I think the reason is that where a user does not exist search has to be used to verify that it does not exist prior to adding the user. Using Yourkit I see about 55% of the time spent in the Lucene index performing a query triggered in UserManagerImpl.getUserNode(userId).

I assume that this query is needed because user nodes can exist in many places in in the repo under /rep:security/rep:authorizables so the path cannot be inferred from the username ?

Was there some design criteria that made this necessary, and is there any way to avoid the search on create.

I think this search must happen in every session that hasn't seen the user before which must be slowing almost everything where these isnt a pooled session down ?

Ian

On 5 Mar 2010, at 11:08, Ian Boston wrote:

> Hi,
> I am trying to optimise a bit of code that adds users to jackrabbit, I have a lot of users to load.
> 
> Are these times unreasonable for Jackrabbit.
> 
> It takes 15ms to get the user with a call like 
> 
>      String principalName = "newUserId"
>      Authorizable authorizable = userManager.getAuthorizable(principalName);
> 
> user does not exist.
> 
> It takes 44ms to create a user with a call like
> 
>        User user = userManager.createUser(principalName, pass, p, prefix);
> 
> prefix is something line "aa/bb/cc/dd"
> 
> aa,bb,cc,dd all exist, and should be in memory and dd has < 255 children
> 
> I also create about 5 other nodes add some simple ACL's and the whole operation takes upto 250ms.
> 
> I am a bit worried, since on in other RDBMS based  systems, that this will be compared against, this sort of operation is faster,
> 
> deployers need to add in the range 10K to 4M, at the top end thats going to take11 days to load if there is no slow down.
> 
> Any suggestions how to make node creation faster?
> Is there some magic trick that I am missing?
> 
> Jackrabbit 1.6
> Java 1.6
> OOTB standard config with Derby as a back end.
> MacBook Pro, 2.53 Ghz Intel Core 2 Duo 4GB DDR3
> 
> 
> Thanks
> Ian