Grouper Users - Open Discussion List

[Michael R Gettes <>]

It’s not that I think base.properties needs to change - i offer my experience 
to be mixed with the experiences of others to help make an informed choice.  
An option might be to offer some recipes in the config files - commented out 
- and who contributed the recipe.  Why the config file?  Well, there’s 
already a bit of documentation in the config files and, unfortunately, 
googling doesn’t necessarily help find everything.

I also have a recipe for configuring InnoDB on MySQL that appears to be 
working quite well based on the experiences of googling and applying ideas of 
various others.  I wish MySQL would include such recipes in their configs so 
MySQL would just work.

I hope this helps.

/mrg

> On Nov 2, 2016, at 12:39 AM, Hyzer, Chris 
> <>
>  wrote:
> 
> We fixed this in august and it was supposed to be in a patch but it appears 
> a patch was never made.  Sorry about that.
> 
> https://bugs.internet2.edu/jira/browse/GRP-1365
> 
> I just made one (patch 30).  Yes, now that loader jobs are threaded, it 
> uses more connections, finishes faster.
> 
> Michael / Scott (not an "Office" reference) or anyone, if you think the 
> params in the default need to be changed let me know
> 
> https://github.com/Internet2/grouper/blob/GROUPER_2_3_BRANCH/grouper/conf/grouper.hibernate.base.properties
> 
> Thanks,
> Chris
> 
> -----Original Message-----
> From: 
> 
>  
> [mailto:]
>  On Behalf Of Michael R. Gettes
> Sent: Tuesday, November 01, 2016 7:35 PM
> To: Scott Koranda 
> <>
> Cc: grouper-users 
> <>
> Subject: Re: [grouper-users] exhausting hibernate JDBC connection pool
> 
> Scott, while I can’t answer the why questions you ask I can share my recipe 
> after spending some time with Shilen at TechX and pushed me to really 
> figure this out.  Here is what we use to manage about 64K course groups for 
> last, current and next semester.  I run 2 loader jobs every morning and 
> processing takes about 1 hour on a 2 CPU VM with 12G (we run loader jobs on 
> 2 instances of Grouper against an Oracle DB on SSD disk).  I hope this 
> helps.
> 
> hibernate.c3p0.max_size = 40
> hibernate.c3p0.min_size = 5
> hibernate.c3p0.idle_test_period = 300
> hibernate.c3p0.timeout = 1800
> hibernate.c3p0.max_statements = 50
> hibernate.c3p0.acquireIncrement = 1
> hibernate.c3p0.checkoutTimeout=60000
> 
> /mrg
> 
>> On Nov 1, 2016, at 17:33, Scott Koranda 
>> <>
>>  wrote:
>> 
>> Hi,
>> 
>> I am helping manage a large Grouper deployment that manages a
>> number of groups for a campus, including groups that represent
>> course offerings ("course groups").
>> 
>> November 1 marks a "transition" from one set of course groups
>> to another. The result is a large amount of work for 3 loader
>> jobs that manage the course groups.
>> 
>> In the past earlier versions of Grouper, all running on the
>> same VM (no changes in the available resources have happened),
>> have ridden out this transition without issue.
>> 
>> Today, however, using Grouper version 2.3 was different. We
>> saw that the loader consumed all of the available JDBC
>> connections. The default was set in
>> grouper.hibernate.base.properties:
>> 
>> hibernate.c3p0.max_size = 16
>> 
>> This caused a number of loader jobs, including maintenance
>> jobs like CHANGE_LOG_changeLogTempToChangeLog, to be starved.
>> 
>> Stopping the loader and changing to
>> 
>> hibernate.c3p0.max_size = 32
>> 
>> made the issue go away.
>> 
>> My question is what might be different in Grouper version 2.3
>> that caused all of the JDBC connections to be consumed where
>> previously they were not? 
>> 
>> I am wondering if code has changed that causes the Grouper
>> loader to be more "aggressive" in how it processes "a lot of
>> work"?
>> 
>> I also note that we did not see this issue on the dev system.
>> That VM, however, has less CPU cores and memory available to
>> it, so I am wondering if that constrained the loader in some
>> way and prevented it from scaling up and exhausting the JDBC
>> connections?
>> 
>> I appreciate any input the Grouper team has.
>> 
>> Thanks,
>> 
>> Scott K
>