TCP keepalives - possible discrepancy with libpq/pg-native

[motiz88]

I'm just beginning to investigate this issue, on a hunch that it is the cause of a problem I'm having in a pg-then-based app (if the network fails during a query, it hangs).

First, I don't see that node-postgres supports the keepalives connection option in any way. Though I don't know if that's intended.

Second, libpq (and therefore pg-native?) enables TCP keepalives by default, and thus should be able to detect a dropped connection; whereas AFAICT Node's net.Socket (and therefore pg) does not. So, with the same connection string that doesn't explicitly set keepalives, different behaviors would be observed with pg vs. pg-native.

Anyone care to comment/verify/enlighten? :)

[NumminorihSF]

I doesn't know, as it is in pg-native, but there is such problem at pg. Also, we have pool leaking, because node doesn't know, that connection is destroyed. I saw this behavior at our project, to start work with postgresql again we should restart application or add some logic with timeouts and wrap sql calls.

I made a PR#913 to add heartbeat to connection and get reaction o destroing of TCP-connection, but have no answer yet.

[NumminorihSF]

I found socket.setKeepAlive([enable][, initialDelay]) method at nodejs socket. It let you know, then socket is destroyed, but it doesn't raise an error if socket is really destroyed.

UPDATE: but it send an error to callback function. I'll update an PR.