""Antti Nivala"" ...

> On a production server, we ran into a situation where all attempts to
> start a new transation returned the following error:
>
> "Transaction start failed. deadlock"
>
> The server was running Firebird 2.0.3.12981 (embedded). Our server
> process has several background threads, and all of them had started to
> fail with this error at the start of each operation.

Is this threads share the same connection or its have own connections ?

> I cannot say
> whether or not there were other transactions in progress. It is
> certainly possible that there were, but I do not understand how it
> became impossible to start _any_ new transactions. The new transactions
> would have been normal "repeatable read" transactions. The error was
> returned from the "start transaction" call. There was no SQL statement
> involved yet (other than "BEGIN TRANSACTION" or something like that).
>
> Since this is a production server, the server process had to be shut
> down and restarted pretty quickly and we could not gather any additional
> information for troubleshooting. After restarting the process,
> everything was fine again.
>
> I'd be interested in knowing what are the likely causes of this kind of
> problem. If the server is running low on memory, can this occur? Or,
> does this indicate a bug in Firebird? Any such known bugs? Something
> else?

Could you post database header statitstics here ? Any chance automatic
sweep was running at time this error's raised ? Does isc_start_transaction
set tra_handle into not NULL value ?

Regards,
Vlad