Re: help needed: hsqldb shutdown race condition, how to fix without deadlock

On Tue, Jun 02, 2015 at 03:12:51PM +0200, Noel Grandin wrote:
> On 2015-06-02 03:07 PM, Lionel Elie Mamane wrote:
> > I found a race condition between

> > 1) connectivity::hsqldb::ODriverDelegator::flushConnections()
> >     (in file connectivity/source/drivers/hsqldb/HDriver.cxx around line 620)
> >     walking over m_aConnections

> > 2) connectivity::hsqldb::ODriverDelegator::disposing()
> >     (same file around line 565)
> >     calling

> >          m_aConnections.clear();
> >          TWeakPairVector().swap(m_aConnections);

> >     which invalidates the iterators used by flushConnection()

> > Plus also probably, race condition between flushConnections() calling
> > flush() and disposing() disposing the underlying connection (although
> > this is probably handled correctly by the exception catching...).

> I would suggest something like this in ODriverDelegator::flushConnections():

>     take_mutex();
>     std::vector<Connection> tmp = m_aConnections;
>     release_mutex();
>     for (Connection conn : tmp)
>         conn.flush();

> so that the mutex is not held during the problematic phase

That makes sense. Will try that. Thanks!

-- 
Lionel