[elephant-devel] Re: BT, concurrency test cases, Postmodern backend changes

Leslie P. Polzer leslie.polzer at gmx.net
Thu Mar 6 14:36:54 UTC 2008


> can you please describe changes you've made so i will not occasionally omit
> them?

Txn changes:

1) Deadlock handling: when a deadlock is detected, one of the txns involved
   will be aborted. Change: catch error and retry the txn instead.

2) Honor the RETRIES parameter for txn restarts.

3) Add warnings for some situations.

What were you working on?


> what is this savepoint thing? i see lots of lines commented out, it was just
> some experimentations?

There were a lot of changes, but none of them involve savepoints yet
(those that are commented out are where they should finally be, but
I haven't got around to get it working smoothly yet).

The savepoints can be used to simulate nested transactions (only roll
back to beginning of innermost txn, not fully), and they might
be helpful in handling the "prepared statement already exists"
problem.


> what's rationale behind dbpm-error, can't we just catch
> cl-postgres:database-error in execute-transaction?

Now that you say it I wonder myself why I didn't do that.
In all likelihood this should be changed.

  Leslie




More information about the elephant-devel mailing list