Unlocking logs when a POSTGRESQL client disconnects abnormally

Use this forum for questions and answers regarding PostGreSQL and the PGDBE.
Message
Author
Diego Euri Almanzar
Posts: 155
Joined: Thu Nov 05, 2020 10:51 am
Location: DOMINICAN REPUBLIC

Unlocking logs when a POSTGRESQL client disconnects abnormally

#1 Post by Diego Euri Almanzar »

I have started my POSGRESQL jobs recently. I was able to start the POSTGRESQL session and use the PGDBE driver, without any problem. But, I had an error with an undefined variable, while running the program, and exited abruptly with Alt+C. Since then, some records, from some tables, have been locked permanently. Even if you turn off the computer, when you turn it on again, the records are still locked.

Is there a way to prewrite, or unlock, the POSTGRESQL session, so that the records are unlocked?

When the client application aborts its connection to the server, or the connection to the application is abnormally terminated or stopped, ADS resets all locks. The default Advantage Database Server client timeout is 120 seconds. Since POSTGRESQL is more powerful, it is supposed to have this kind of reset or unlock tool on abnormality.

Someone who can help me, please?

Best regards

k-insis
Posts: 94
Joined: Fri Jan 28, 2011 4:07 am

Re: Unlocking logs when a POSTGRESQL client disconnects abnormally

#2 Post by k-insis »

Pgdbe (when using 'upsize' tool) uses a addition to every table, namely :

__deleted boolean NOT NULL DEFAULT false,
__record integer NOT NULL DEFAULT nextval('sometable___record_seq'::regclass),
__rowversion integer NOT NULL DEFAULT 0,
__keyversion integer NOT NULL DEFAULT 0,
__lock_owner integer NOT NULL DEFAULT 0,

And without data it might be that table.__lock_owner has something written inside it apart from 0 (zero).

Use pgadmin to check contents of those fields at problematic tables.


Diego Euri Almanzar wrote: Sat Apr 09, 2022 7:49 pm I have started my POSGRESQL jobs recently. I was able to start the POSTGRESQL session and use the PGDBE driver, without any problem. But, I had an error with an undefined variable, while running the program, and exited abruptly with Alt+C. Since then, some records, from some tables, have been locked permanently. Even if you turn off the computer, when you turn it on again, the records are still locked.

Is there a way to prewrite, or unlock, the POSTGRESQL session, so that the records are unlocked?

When the client application aborts its connection to the server, or the connection to the application is abnormally terminated or stopped, ADS resets all locks. The default Advantage Database Server client timeout is 120 seconds. Since POSTGRESQL is more powerful, it is supposed to have this kind of reset or unlock tool on abnormality.

Someone who can help me, please?

Best regards

User avatar
Tom
Posts: 1165
Joined: Thu Jan 28, 2010 12:59 am
Location: Berlin, Germany

Re: Unlocking logs when a POSTGRESQL client disconnects abnormally

#3 Post by Tom »

You're right, Diego. If the app is killed, the lock stays. As k-insis mentioned, there is a table row "__lock_owner" in every table, which is 0 (zero) if a record is not locked. If it contains an integer value, the record is (and stays) locked. This value is not removed if an app doesn't close and quit normally. I reported the problem to Alaska Software, but I think, this will be hard to fix. The ADS keeps locks in memory, not in the tables, so if a session times out, all locks for tables used by a session get invalid. Since the PGDBE stores the lock in the table, there is much more to do.
Best regards,
Tom

"Did I offend you?"
"No."
"Okay, give me a second chance."

User avatar
Tom
Posts: 1165
Joined: Thu Jan 28, 2010 12:59 am
Location: Berlin, Germany

Re: Unlocking logs when a POSTGRESQL client disconnects abnormally

#4 Post by Tom »

Code: Select all

UPDATE public.<mytablename> SET __lock_owner = 0
in pgAdmin or from your app (maybe in service routines, after DbPack) will solve temporarly.
Best regards,
Tom

"Did I offend you?"
"No."
"Okay, give me a second chance."

Diego Euri Almanzar
Posts: 155
Joined: Thu Nov 05, 2020 10:51 am
Location: DOMINICAN REPUBLIC

Re: Unlocking logs when a POSTGRESQL client disconnects abnormally

#5 Post by Diego Euri Almanzar »

Excellent information. Anyway, I'll write to Alaska Software, and let them know what answer they give me.

Thank you, K-INSIS and TOM.

k-insis
Posts: 94
Joined: Fri Jan 28, 2011 4:07 am

Re: Unlocking logs when a POSTGRESQL client disconnects abnormally

#6 Post by k-insis »

Can you copy/paste reply by AS ? It would be interesting if and if, what they say about that (quite big) issue.

Unless they use some additional timestamp and limit 'lock' on limited timeslot, or even use another table with information.... . Concept of record locking in sql is quite different to file based in dbf .

Diego Euri Almanzar wrote: Mon Apr 11, 2022 9:34 pm Excellent information. Anyway, I'll write to Alaska Software, and let them know what answer they give me.

Thank you, K-INSIS and TOM.

User avatar
Tom
Posts: 1165
Joined: Thu Jan 28, 2010 12:59 am
Location: Berlin, Germany

Re: Unlocking logs when a POSTGRESQL client disconnects abnormally

#7 Post by Tom »

Alaska opened up a PDR for the issue: https://www.alaska-software.com/scripts ... PDRID=7491
Best regards,
Tom

"Did I offend you?"
"No."
"Okay, give me a second chance."

Diego Euri Almanzar
Posts: 155
Joined: Thu Nov 05, 2020 10:51 am
Location: DOMINICAN REPUBLIC

Re: Unlocking logs when a POSTGRESQL client disconnects abnormally

#8 Post by Diego Euri Almanzar »

Hello,Tom

well, i'm worried. Because I went to the link you gave me, and Alaska explains that she has no solution for that problem.

In any case, I will send an email to Alaska, since in a year of subscription that corresponds to me, I have only sent an email. Maybe a miracle will happen and they will give me another kind of answer.

Thank you.

User avatar
Tom
Posts: 1165
Joined: Thu Jan 28, 2010 12:59 am
Location: Berlin, Germany

Re: Unlocking logs when a POSTGRESQL client disconnects abnormally

#9 Post by Tom »

Hi, Diego.
Because I went to the link you gave me, and Alaska explains that she has no solution for that problem.
The problem is a part of the "Defect Tracking System" and became a "Problem Report":

https://www.alaska-software.com/support/kbase-old.cxp

There is no solution yet, since they just started working on a solution. If you face this problem, replace __lock_owner from inside pgAdmin or using the SQL statement I posted before. We have huge PGDBE/ISAM-apps (late beta) running in a productive environment at several customer sites, and we didn't face this problem in reality. This maybe by design - we never lock records while data is in revision, but only if the data is written. This reduces lock times to some seconds a day. And a "normal" runtime error ensures an unlock. The deadlock problem occurs only if a fatal error or an ALT-C come up. You can switch off ALT-C for your app if you want (SetCancel(.F.)).

I'm quite sure this problem will be fixed in the next update.
Best regards,
Tom

"Did I offend you?"
"No."
"Okay, give me a second chance."

k-insis
Posts: 94
Joined: Fri Jan 28, 2011 4:07 am

Re: Unlocking logs when a POSTGRESQL client disconnects abnormally

#10 Post by k-insis »

Locking just before change and unlocking right after update is more sensible solution, including for regular dbf / it spares server resources.

Same situation with 'hanging' locks happens if connection between client and server is lost while app is running with active locks.

But if AS might move this functionality to server side set of stored procedures which would keep a list of locked records inside access table depending on active connections PIDs (from "SELECT * FROM pg_stat_activity" ) and pgsql via rlock(), dbunlock(), flock(), etc would just request status change, status check and status remove for table.___recno by executing those stored procedures with parameters.

A single stored procedure might just be pushed by pg_cron job once per minute to remove all orphaned indices for locks from table if there is no longer active connection for a certain rlock() or flock() .

Sure a bit more traffic and some server cpu time/iops, but it would not depend on client to properly function anymore.

But hard to say what they will do at end.
Tom wrote: Wed Apr 13, 2022 12:07 am Hi, Diego.
Because I went to the link you gave me, and Alaska explains that she has no solution for that problem.
The problem is a part of the "Defect Tracking System" and became a "Problem Report":

https://www.alaska-software.com/support/kbase-old.cxp

There is no solution yet, since they just started working on a solution. If you face this problem, replace __lock_owner from inside pgAdmin or using the SQL statement I posted before. We have huge PGDBE/ISAM-apps (late beta) running in a productive environment at several customer sites, and we didn't face this problem in reality. This maybe by design - we never lock records while data is in revision, but only if the data is written. This reduces lock times to some seconds a day. And a "normal" runtime error ensures an unlock. The deadlock problem occurs only if a fatal error or an ALT-C come up. You can switch off ALT-C for your app if you want (SetCancel(.F.)).

I'm quite sure this problem will be fixed in the next update.

Post Reply