I have a code change that resolves this issue in a flexible way, before I check it in can anyone see any flaws in it?
I decided to allow the first character of the db_url to be set to ‘*’ to indicate that non-pooling is required. In this way the pooling can be controlled from the config file giving extra flexibility. I have added a new variable poolid to db_id struct. By default this is zero, however in the case that non-pooling is required the poolid is set to a unique value within that pid ensuring the db_id struct is unique and non pooled.
Any comments? Good or bad
The code changes are fairly minimal and limited to lib/srdb1: db_id.c, db_id.h, db.c
Paul
From: Daniel-Constantin Mierla Sent: Tuesday, December 13, 2011 3:53 PM To: Development mailing list of the sip-router project Cc: Paul Pankhurst Subject: Re: [sr-dev] db_fetch_next problem with Postgres?
On 12/13/11 4:16 PM, Paul Pankhurst wrote: Thanks for that Daniel.
I modified the db_url for 2nd db so it used the actual IP address of my db rather than localhost, nice idea, indeed -- perhaps can be extended to add some custom hostnames pointing to same in local computer's /etc/hosts, in case one would need more dedicated connection. But it would be better if APi gets extended.
Cheers, Daniel
hence it didn’t get pooled, and everything magically worked.
I think I’ll extend the db api, it doesn’t look too difficult to do.
Paul
From: Daniel-Constantin Mierla Sent: Tuesday, December 13, 2011 12:27 PM To: Development mailing list of the sip-router project Cc: Paul Pankhurst Subject: Re: [sr-dev] db_fetch_next problem with Postgres?
Hello,
On 12/13/11 1:08 PM, Paul Pankhurst wrote: should have sent this to list – sorry
From: Paul Pankhurst Sent: Tuesday, December 13, 2011 12:03 PM To: miconda@gmail.com Subject: Re: [sr-dev] db_fetch_next problem with Postgres?
Hi Daniel,
I have a bit more information...
the outline pseudo code for what I am doing looks like this (removing non relevant stuff)
db_fetch_query( xcap_db )
do {
query( another_2nd_db )
} while ( db_fetch_next( xcap_db) )
it is the query to the second database that breaks everything...if I remove it the problem goes away and db_fetch_next works as expected. The access to the 2nd database uses completely different db_func_t and db1_cont_t structures to the xcap_db
if it is the same db_url value, then it results in same db1_con_t pointers, since the postgres has connection pool inside and matches an old connection for same db_url when initializing the second one.
Some ideas: - extend db api with a new function init_new that will always return a new connection -- requires coding in db modules - make another database where you add a view to the table in initial database and work with it for second connection -- should work without touching kamailio code
Cheers, Daniel
I’ll try running it with extra debug shortly to see if I can see what’s happening
Paul
From: Daniel-Constantin Mierla Sent: Tuesday, December 13, 2011 11:10 AM To: Development mailing list of the sip-router project Cc: Paul Pankhurst Subject: Re: [sr-dev] db_fetch_next problem with Postgres?
Hello,
On 12/13/11 11:40 AM, Paul Pankhurst wrote: I am having problems with the db_fetch_next routine and Postgres.
I can’t see any problem with my code that calls it, and as far as I can tell the problem appears to be in the db_postgres_fetch_result routine. On the first invocation of db_postgres_fetch_result it allocates a result structure and gives me the first 5 rows of the 198 that match my query after calling db_postgres_convert_rows. On the 2nd invocation it fails to retrieve the next 5 rows because CON_RESULT(_con) is NULL ! Anyone got any suggestions? can you doublecheck and see if you don't call free_result() before fetching the next rows? If you run with debug=3, if there is a free_result, you should see the log message from:
LM_DBG("PQclear(%p) result set\n", CON_RESULT(_con))
Cheers, Daniel
