pool configuration could be an api call (rather than set in layer XML)

[artemp]

After the bugfix as part of #855, it is now possible for servers that use more than 10 rendering threads to max out the default max_size of 10 for the postgres pool, leading to undefined behavior.

So, even after this is fixed (#862), it would still be more ideal for the calling application to be able to set the pool max_size, because this is actually a global setting and only the calling application that knows the # of potential threads that may operate concurrently (not the stylesheet author).

[springmeyer]

if we move on this, then this hardcoded default should likely be surfaced in tilemill: https://github.com/mapbox/tilemill/blob/master/commands/export.bones#L206-L211

[strk]

Agreed that a direct API to the pooler would be useful. For example someone might want to drop all unused connections periodically, or at a given point in time. At least, that's what I currently need to do, to avoid ending up out of file descriptor (ouch..)

[strk]

I've spent some time looking at node-pool, a generic pooler for node.js and I like a couple of behavior that I think would be nice to have in mapnik too:

1. Async acquisition of a resource from the pool. When no connections are available and no more connections can be created (limit hit) the requests queued up waiting for a resource to become available.
2. Expiration of idle connections. After some time that a connection is in the pool with nobody requesting it, it's shut down and removed from the queue.