Bug 617623

Summary:	SSSD suffers from serious performance issues on initgroups calls
Product:	Red Hat Enterprise Linux 6	Reporter:	Stephen Gallagher <sgallagh>
Component:	sssd	Assignee:	Stephen Gallagher <sgallagh>
Status:	CLOSED CURRENTRELEASE	QA Contact:	Chandrasekar Kannan <ckannan>
Severity:	high	Docs Contact:
Priority:	low
Version:	6.0	CC:	benl, dpal, grajaiya, jgalipea, pere, trondham, werner
Target Milestone:	rc
Target Release:	---
Hardware:	All
OS:	Linux
Whiteboard:
Fixed In Version:	sssd-1.2.1-23.el6	Doc Type:	Bug Fix
Doc Text:		Story Points:	---
Clone Of:		Environment:
Last Closed:	2010-11-10 21:40:05 UTC	Type:	---
Regression:	---	Mount Type:	---
Documentation:	---	CRM:
Verified Versions:		Category:	---
oVirt Team:	---	RHEL 7.3 requirements from Atomic Host:
Cloudforms Team:	---	Target Upstream Version:
Embargoed:
Bug Depends On:
Bug Blocks:	579775

Description Stephen Gallagher 2010-07-23 14:59:23 UTC

Description of problem:
Users who are a member of one or more groups containing large numbers of members end up with unacceptable wait times (or timeouts) performing initgroups calls.

Since initgroups calls are always performed on login, this is causing unacceptably long login waits.


Version-Release number of selected component (if applicable):
sssd-1.2.1-21.el6

How reproducible:
Every time

Steps to Reproduce:
1. Create 10,000 users in LDAP.
2. Create a group that contains all of these users
3. Attempt to log in as one of these users.

  
Actual results:
There is a very long wait time to log in (measured in seconds, if not tens of seconds depending on system I/O performance)


Expected results:
Login time should be reasonable ( < 5 seconds)

Additional info:
This is happening because we tried to be too clever with the initgroups calls. When a user logs in, we attempt to completely refresh all entries associated with all groups the user is a member of. This worked fine in our small test environments, but we didn't adequately test this for scalability. Our customers are now doing this for us and are not pleased with the results.

Comment 1 Stephen Gallagher 2010-07-30 20:01:44 UTC

Testers: you can compile and run the following C program and run it with the 'time' command

getgroups.c:

#include <sys/types.h>
#include <unistd.h>
#include <grp.h>

int main(int argc, char **argv)
{
    initgroups(argv[1], 0);

    return 0;
}



Compile and run:
gcc -o initgroups.test getgroups.c
time ./initgroups.test <username>


Prior to this patch, time would have reported 3-5s for users in groups of ~300-500 users. After this patch it will average < 1s.

Please test this on users of varying sized groups and numbers of groups.

Comment 3 Gowrishankar Rajaiyan 2010-08-30 12:50:25 UTC

Setup:

1. Created a group bosgroup with 500 users, Group1 with 5000 users and mumgroup with 10000 users.
2. bos 500 is a member of Group1 and mumgroup.
3. pnq5000 is a member of mumgroup.
4. mum9999 is a member on mumgroup.

[root@rhel6snap11 ~]# time ./initgroups.test bos500

real	0m0.173s
user	0m0.000s
sys	0m0.001s
[root@rhel6snap11 ~]# time ./initgroups.test pnq5000

real	0m0.226s
user	0m0.001s
sys	0m0.000s
[root@rhel6snap11 ~]# time ./initgroups.test mum9999

real	0m0.272s
user	0m0.000s
sys	0m0.001s

Verified: sssd-1.2.1-26.el6.

Comment 4 Petter Reinholdtsen 2010-09-11 06:03:37 UTC

Where can I find the patch used to fix this issue?

Comment 5 Stephen Gallagher 2010-09-13 10:59:59 UTC

(In reply to comment #4)
> Where can I find the patch used to fix this issue?

http://git.fedorahosted.org/git/?p=sssd.git;a=commit;h=24a5809cbc96d10001535999a292cb26242ad01d

However, as is being discussed right now at https://fedorahosted.org/pipermail/sssd-devel/2010-September/004540.html we've discovered that this patch has introduced some issues with group resolution under certain circumstances.

Comment 6 releng-rhel@redhat.com 2010-11-10 21:40:05 UTC

Red Hat Enterprise Linux 6.0 is now available and should resolve
the problem described in this bug report. This report is therefore being closed
with a resolution of CURRENTRELEASE. You may reopen this bug report if the
solution does not work for you.