You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@directory.apache.org by Ca...@ibs-ag.com on 2012/10/10 04:55:55 UTC

build monday from trunk very slow for ldif import

Hi, I'm still trying finish the testing of all the Password policy work Karin did over the weekend but I have another issue that's come up.  Ldif imports are extremely slow.

During our testing, we often delete the entire partition directory to start fresh. When the server starts, it lays down the partition and .db files as defined config.ldif.

Anyway, we used to import (via studio) an ldif file with 80k entries and it would load about 90 entries per second.  That was great!
With this build it's going at about 4-5 entries per second.

We noticed is that previously, the partition .db files would not change (on disk) until after the ldif import was complete.
Then when we stopped the server, it was like the entire import got flushed to the disk at once. The files would go from 20K to 400MB.
With this build, it seems to be updating the files as it goes. Could this be the reason?

Also, this is the first build I noticed the .lg files in the partition directory. I think they're there for journaling but don't know if that's an option something new?

We removed all the password policy Attributes from my ldif file thinking that was slowing it down but it's essentially the same performance.
Below is my partition and all the indexes are set like the one I included. Any changes that would affect this in the last few weeks.  Anyone else seeing this? Thanks!

dn: ads-partitionId=cpro,ou=partitions,ads-directoryServiceId=default,ou=config
objectclass: top
objectClass: ads-base
objectclass: ads-partition
objectclass: ads-jdbmPartition
ads-indexes: apacheRdn
ads-indexes: apacheSubLevel
ads-indexes: apachePresence
ads-indexes: apacheOneLevel
ads-indexes: apacheOneAlias
ads-indexes: apacheSubAlias
ads-indexes: apacheAlias
ads-indexes: entryCSN
ads-indexes: krb5PrincipalName
ads-indexes: objectClass
ads-indexes: ou
ads-indexes: uid
ads-indexes: employeeNumber
ads-indexes: displayName
ads-indexes: cn
ads-indexes: mail
ads-indexes: roomNumber
ads-indexes: pwdPolicySubEntry
ads-indexes: member
ads-indexes: description
ads-indexes: givenName
ads-indexes: sn
ads-indexes: administrativeRole
ads-partitionSuffix: o=cpro
ads-jdbmpartitionoptimizerenabled: TRUE
ads-partitioncachesize: 100
ads-partitionsynconwrite: TRUE
ads-partitionid: cpro
ads-enabled: TRUE

#index example, they're all like this..HasReverse=FALSE
dn: ads-indexAttributeId=uid,ou=indexes,ads-partitionId=cpro,ou=partitions,ads-directoryServiceId=default,ou=config
ads-indexattributeid: uid
ads-indexHasReverse: FALSE
ads-indexcachesize: 100
objectclass: ads-index
objectclass: ads-jdbmIndex
objectclass: ads-base
objectclass: top
ads-enabled: TRUE

Re: build monday from trunk very slow for ldif import

Posted by Kiran Ayyagari <ka...@apache.org>.
we should also change the script to start JVM with -Xms512m and -server options

On Wed, Oct 10, 2012 at 4:22 PM, Emmanuel Lécharny <el...@gmail.com> wrote:
> Le 10/10/12 12:03 PM, Carlo.Accorsi@ibs-ag.com a écrit :
>
>> ads-partitionSyncOnWrite=FALSE did the trick!   Back to 80 adds/sec, Thank
>> you!!
>
> Cool. We may have to set this flag to false by default...
>
>
>
> --
> Regards,
> Cordialement,
> Emmanuel Lécharny
> www.iktek.com
>



-- 
Kiran Ayyagari
http://keydap.com

Re: build monday from trunk very slow for ldif import

Posted by Emmanuel Lécharny <el...@gmail.com>.
Le 10/10/12 12:03 PM, Carlo.Accorsi@ibs-ag.com a écrit :
> ads-partitionSyncOnWrite=FALSE did the trick!   Back to 80 adds/sec, Thank you!!
Cool. We may have to set this flag to false by default...


-- 
Regards,
Cordialement,
Emmanuel Lécharny
www.iktek.com


RE: build monday from trunk very slow for ldif import

Posted by Ca...@ibs-ag.com.
ads-partitionSyncOnWrite=FALSE did the trick!   Back to 80 adds/sec, Thank you!!


Regards,
Carlo Accorsi


-----Original Message-----
From: Emmanuel Lécharny [mailto:elecharny@gmail.com] 
Sent: Wednesday, October 10, 2012 3:16 AM
To: users@directory.apache.org
Subject: Re: build monday from trunk very slow for ldif import

Le 10/10/12 4:55 AM, Carlo.Accorsi@ibs-ag.com a écrit :
> Hi, I'm still trying finish the testing of all the Password policy work Karin did over the weekend but I have another issue that's come up.  Ldif imports are extremely slow.
>
> During our testing, we often delete the entire partition directory to start fresh. When the server starts, it lays down the partition and .db files as defined config.ldif.
>
> Anyway, we used to import (via studio) an ldif file with 80k entries and it would load about 90 entries per second.  That was great!
> With this build it's going at about 4-5 entries per second.
Hmmm. This is very slow. How many indexed attributes do you have ?

I have done some tests locally, and I'm able to get up to 200 add/s, but with a simpler entry.
>
> We noticed is that previously, the partition .db files would not change (on disk) until after the ldif import was complete.
> Then when we stopped the server, it was like the entire import got flushed to the disk at once. The files would go from 20K to 400MB.
> With this build, it seems to be updating the files as it goes. Could this be the reason?
It may be because we flush on disk for every single write. You could turn the ads-partitionSyncOnWrite flag to FALSE, so that the data are flush only ever ads-dsSyncPeriodMillis (default to 15 seconds)
>
> Also, this is the first build I noticed the .lg files in the partition directory. I think they're there for journaling but don't know if that's an option something new?
It's a JDBM file which is created beside the db files, AFAIR. I have to double check that.
>
> We removed all the password policy Attributes from my ldif file thinking that was slowing it down but it's essentially the same performance.
> Below is my partition and all the indexes are set like the one I included. Any changes that would affect this in the last few weeks.  Anyone else seeing this? Thanks!
>
> dn: 
> ads-partitionId=cpro,ou=partitions,ads-directoryServiceId=default,ou=c
> onfig
> objectclass: top
> objectClass: ads-base
> objectclass: ads-partition
> objectclass: ads-jdbmPartition
> ads-indexes: apacheRdn
> ads-indexes: apacheSubLevel
> ads-indexes: apachePresence
> ads-indexes: apacheOneLevel
> ads-indexes: apacheOneAlias
> ads-indexes: apacheSubAlias
> ads-indexes: apacheAlias
> ads-indexes: entryCSN
> ads-indexes: krb5PrincipalName
> ads-indexes: objectClass
> ads-indexes: ou
> ads-indexes: uid
> ads-indexes: employeeNumber
> ads-indexes: displayName
> ads-indexes: cn
> ads-indexes: mail
> ads-indexes: roomNumber
> ads-indexes: pwdPolicySubEntry
> ads-indexes: member
> ads-indexes: description
> ads-indexes: givenName
> ads-indexes: sn
> ads-indexes: administrativeRole
> ads-partitionSuffix: o=cpro
> ads-jdbmpartitionoptimizerenabled: TRUE
> ads-partitioncachesize: 100
> ads-partitionsynconwrite: TRUE
> ads-partitionid: cpro
> ads-enabled: TRUE
>
> #index example, they're all like this..HasReverse=FALSE
> dn: 
> ads-indexAttributeId=uid,ou=indexes,ads-partitionId=cpro,ou=partitions
> ,ads-directoryServiceId=default,ou=config
> ads-indexattributeid: uid
> ads-indexHasReverse: FALSE
> ads-indexcachesize: 100
> objectclass: ads-index
> objectclass: ads-jdbmIndex
> objectclass: ads-base
> objectclass: top
> ads-enabled: TRUE
>


--
Regards,
Cordialement,
Emmanuel Lécharny
www.iktek.com


Re: build monday from trunk very slow for ldif import

Posted by Emmanuel Lécharny <el...@gmail.com>.
Le 10/10/12 4:55 AM, Carlo.Accorsi@ibs-ag.com a écrit :
> Hi, I'm still trying finish the testing of all the Password policy work Karin did over the weekend but I have another issue that's come up.  Ldif imports are extremely slow.
>
> During our testing, we often delete the entire partition directory to start fresh. When the server starts, it lays down the partition and .db files as defined config.ldif.
>
> Anyway, we used to import (via studio) an ldif file with 80k entries and it would load about 90 entries per second.  That was great!
> With this build it's going at about 4-5 entries per second.
Hmmm. This is very slow. How many indexed attributes do you have ?

I have done some tests locally, and I'm able to get up to 200 add/s, but 
with a simpler entry.
>
> We noticed is that previously, the partition .db files would not change (on disk) until after the ldif import was complete.
> Then when we stopped the server, it was like the entire import got flushed to the disk at once. The files would go from 20K to 400MB.
> With this build, it seems to be updating the files as it goes. Could this be the reason?
It may be because we flush on disk for every single write. You could 
turn the ads-partitionSyncOnWrite flag to FALSE, so that the data are 
flush only ever ads-dsSyncPeriodMillis (default to 15 seconds)
>
> Also, this is the first build I noticed the .lg files in the partition directory. I think they're there for journaling but don't know if that's an option something new?
It's a JDBM file which is created beside the db files, AFAIR. I have to 
double check that.
>
> We removed all the password policy Attributes from my ldif file thinking that was slowing it down but it's essentially the same performance.
> Below is my partition and all the indexes are set like the one I included. Any changes that would affect this in the last few weeks.  Anyone else seeing this? Thanks!
>
> dn: ads-partitionId=cpro,ou=partitions,ads-directoryServiceId=default,ou=config
> objectclass: top
> objectClass: ads-base
> objectclass: ads-partition
> objectclass: ads-jdbmPartition
> ads-indexes: apacheRdn
> ads-indexes: apacheSubLevel
> ads-indexes: apachePresence
> ads-indexes: apacheOneLevel
> ads-indexes: apacheOneAlias
> ads-indexes: apacheSubAlias
> ads-indexes: apacheAlias
> ads-indexes: entryCSN
> ads-indexes: krb5PrincipalName
> ads-indexes: objectClass
> ads-indexes: ou
> ads-indexes: uid
> ads-indexes: employeeNumber
> ads-indexes: displayName
> ads-indexes: cn
> ads-indexes: mail
> ads-indexes: roomNumber
> ads-indexes: pwdPolicySubEntry
> ads-indexes: member
> ads-indexes: description
> ads-indexes: givenName
> ads-indexes: sn
> ads-indexes: administrativeRole
> ads-partitionSuffix: o=cpro
> ads-jdbmpartitionoptimizerenabled: TRUE
> ads-partitioncachesize: 100
> ads-partitionsynconwrite: TRUE
> ads-partitionid: cpro
> ads-enabled: TRUE
>
> #index example, they're all like this..HasReverse=FALSE
> dn: ads-indexAttributeId=uid,ou=indexes,ads-partitionId=cpro,ou=partitions,ads-directoryServiceId=default,ou=config
> ads-indexattributeid: uid
> ads-indexHasReverse: FALSE
> ads-indexcachesize: 100
> objectclass: ads-index
> objectclass: ads-jdbmIndex
> objectclass: ads-base
> objectclass: top
> ads-enabled: TRUE
>


-- 
Regards,
Cordialement,
Emmanuel Lécharny
www.iktek.com