Hello!
Should I really add an index for modifyTimestamp? I got "slapd[4021]: <= bdb_inequality_candidates: (modifyTimestamp) not indexed!" (No discussions on using BDB, please!)
Regards, Ulrich
Ulrich Windl wrote:
Should I really add an index for modifyTimestamp? I got "slapd[4021]: <= bdb_inequality_candidates: (modifyTimestamp) not indexed!" (No discussions on using BDB, please!)
The pure log message does not say much. It depends on the real filters used whether you should index an attribute or not. Hint: You can make *search* performance worse with more indexes.
Ciao, Michael.
P.S.: The log message is not helpful and e.g. with my Æ-DIR it's itself a performance hog (see ITS#7796)
Michael Ströder michael@stroeder.com schrieb am 06.07.2017 um 14:29 in
Nachricht 55cd37d3-e407-877e-01be-096f9f7aacae@stroeder.com:
Ulrich Windl wrote:
Should I really add an index for modifyTimestamp? I got "slapd[4021]: <= bdb_inequality_candidates: (modifyTimestamp) not indexed!" (No discussions
on using
BDB, please!)
The pure log message does not say much. It depends on the real filters used
whether you should index an attribute or not. Hint: You can make *search* performance worse with more indexes.
Actually the accesslog database wanted that index, while no queries (not that I knew) were made on the accesslog database. So any access to it would be some change operation in the main database (that's how accesslog is configures: only log changes). So I don't know the filter.
It may be interesting that I had this configuration for more than a year now, and those messages started to pop up only recently (a month ago or so). The reqMod (largest file) has just about 51 MB in size, and reqStart has about 200kB in size....
At least those messages are gone after having added the index.
Regards, Ulrich
Ciao, Michael.
P.S.: The log message is not helpful and e.g. with my Æ-DIR it's itself a performance hog (see ITS#7796)
Ulrich Windl wrote:
Michael Ströder michael@stroeder.com schrieb am 06.07.2017 um 14:29 in
Nachricht 55cd37d3-e407-877e-01be-096f9f7aacae@stroeder.com:
Ulrich Windl wrote:
Should I really add an index for modifyTimestamp? I got "slapd[4021]: <= bdb_inequality_candidates: (modifyTimestamp) not indexed!" (No discussions
on using
BDB, please!)
The pure log message does not say much. It depends on the real filters used
whether you should index an attribute or not. Hint: You can make *search* performance worse with more indexes.
Actually the accesslog database wanted that index, while no queries (not that I knew) were made on the accesslog database. So any access to it would be some change operation in the main database (that's how accesslog is configures: only log changes). So I don't know the filter.
You own the logs => you're the only one who can find out.
Ciao, Michael.
"Ulrich Windl" Ulrich.Windl@rz.uni-regensburg.de schrieb am 05.07.2017 um
13:59 in Nachricht 595CD493020000A100026F94@gwsmtp1.uni-regensburg.de:
Hello!
Should I really add an index for modifyTimestamp? I got "slapd[4021]: <= bdb_inequality_candidates: (modifyTimestamp) not indexed!" (No discussions on using BDB, please!)
In the meantime I added the index to the main database and to the accesslog database. As the message did not indicate which database it is referring to, I added both. As it turned out the message didn't stop after having added the index to the main database, I guess the accesslog database really needed that index.
FYI: Unfortunately (for unknow reasons) adding that second index took more than 7 minutes, and during that time slapd refused to process any requests (on the replicated servers it took less than one second). The resulting index was 8kB in size, so I guess that shouldn't have taken that long. The effect is being investigated...
Regards, Ulrich
openldap-technical@openldap.org
-
Michael Ströder -
Ulrich Windl