On Thu, 27 Feb 2003 13:05:27 +1100
Philip Yarra <philip(at)utiba(dot)com> wrote:
> I have some ECPG code that is failing intermittently. It's in C, using
> pthreads. There are n threads, each has its own connection to the DB server.
> The code section that fails looks like this:
> log_event(0,SMS_LOG_INFO,tptr->thread_num,"got here 6");
> EXEC SQL execute insert_cache using :session_id, :subscriber_id,
> :service_id, :points_hist
> ory_id, :service_result_code, :source_address, :source_address_type_id,
> , :destination_address_type_id, :log_category_id, :severity_id,
> :transaction_id, :event_loca
> tion, :country_code, :operator_code, :dvr_data;
> log_event(0,SMS_LOG_INFO,tptr->thread_num,"got here 7");
> In a test run with 20 transactions:
> log_event 6 was executed 20 times
> 20 records got inserted in the DB
> log_event 7 was executed 19 times
> Oh yes, the app is compiled using ecpg -t, so it's all auto-committing.
> It looks to me like EXEC SQL is not returning... even though a DB record is
> actually inserted. Problem seems to occur if a new insert is attempted while
> one is still executing. (the actual code section shown here takes just under
> 40 ms - if a request comes within that time, the first insert never returns).
> If the app is run with only one DB thread, it works fine... when run with 3 it
> eventually dies as one thread locks up on the "EXEC SQL execute" call.
> Does anyone have any clues about this? I'm not really sure how to proceed.
> Detail about platform etc below. I can send more detail if required.
> Regards, Philip Yarra.
> DB server: PostgreSQL 7.3.2 on alphaev6-dec-osf4.0f, compiled by GCC 188.8.131.52
> client app: Linux (RH7.3) using gcc 2.96, plus ECPG of course to compile .pgc
> files. Using PostgreSQL 7.3.2 libs.
> The overall app is a little complicated to explain... but basically, each DB
> thread starts, opens its connection to the DB server, EXEC SQL PREPAREs the
> insert_cache variable (one per thread), stores that in a thread-local
> variable and gets ready for insert requests. At each request, it retrieves
> the PREPAREd statement, and EXECUTEs it with the values it was passed. I've
> checked the mem location of the insert_cache per each thread - they're not
> stomping on each other that I can tell. It looks like a threading problem
> with ecpg?
I have not thread programming experience with ecpg. But if you want real nonblocking interface and get a best speed performance
with a lot simultaneos query processing use libpq PQsocket(), PQsendQuery(), PQconsumeInput(), PQbusy function with multiplae connections to PostgreSQL and poll() on connection sockets, including other sockets application handling, to catch up the query results. The idea is to have a free connections stack in a single process application (it may be a single thread). When the application wants to query postgres, it takes first free PGconn* from the stack and sends the query by PQsendQuery without any block of the process. Then poll() is awaiting the result of queries. If you are interest of details mail me. I have used this method and had a best speed of query processing and system resource economy then multithreading (I have tryed it, but with libpq).
mailto: archibald(at)mastak(dot)ru < Archibald >
In response to
pgsql-interfaces by date
|Next:||From: pgsql-bugs||Date: 2003-02-27 11:35:19|
|Subject: Bug #904: Deallocating of prepared statement in ECPG at COMMIT|
|Previous:||From: ash||Date: 2003-02-27 04:40:55|
|Subject: Darwin libpq link failing |