connectionTimeoutMillis doesn't work

[bydga]

Im trying to use the connTimeoutMilis feature (respond error when a connection from a pool cannot be obtained within specified limit) - and it doesnt work.

Here is a minimal example:

{ Pool } = require 'pg'

process.on 'uncaughtException', (err) -> console.log "uncaught", err

pool = new Pool
	connectionString: "tcp://usr:pass@XXX.us-west-2.rds.amazonaws.com/db_name?ssl=true"
	max: 3
	connectionTimeoutMillis: 10000

pool.on 'error', (e, client) -> console.log 'pool.on.err', e

pool.connect (e, client, done) ->
	console.log "first", e
	pool.connect (e, client, done) ->
		console.log "second", e
		pool.connect (e, client, done) ->
			console.log "third", e
			pool.connect (e, client, done) ->
				console.log "fourth", e

output:

$ coffee pg.coffee
first undefined
second undefined
third undefined

and nothing more, no error thrown, no err callback being called..

versions:

$ npm ls | egrep "pg|postgr"
├─┬ pg@7.0.2
│ ├── pg-connection-string@0.1.3
│ ├── pg-pool@2.0.1
│ ├─┬ pg-types@1.12.0
│ │ ├── postgres-array@1.0.2
│ │ ├── postgres-bytea@1.0.0
│ │ ├── postgres-date@1.0.3
│ │ └── postgres-interval@1.1.1
│ ├─┬ pgpass@1.0.2

[brianc]

✅ small self contained steps to reproduce
✅ listing of all the versions in use

Thanks for the nice bug report! I'll look at this ASAP.

[charmander]

@brianc, is this the intended behaviour of connectionTimeoutMillis? i.e. should it only apply when connecting, or should it apply when the pool has reached its limit and the caller has to wait for the next available client as well?

[ghafran]

There is a bug with 7.x where the pooled client never closes, no matter what the connectionTimeoutMillis is set to.

Even after calling done, the connection remains open indefinitely.
this.pool.connect(function (err, client, done) { done(); });

This works fine with 6.1.0. We have reverted back to 6.1.0 until this is resolved because its causing connection leaks.

[charmander]

@ghafran This issue is about connectionTimeoutMillis, not idleTimeoutMillis.

[brianc]

@charmander thanks for drawing my attention back to this one. Got lost in the never-ending issue shuffle. I wanted connectionTimeoutMillis to apply both when initially connecting and waiting for a client from a full pool. The thinking behind that is you're calling connect either time regardless so it should always behave the same way. Likewise pool.query should reject after connectionTimeoutMillis if the timeout is exceeded. I'll look at this right away.

[brianc]

@bydga - My apologies for the delay in responding here to the bug; I get a lot of issues and sometimes they slip off my radar...I've fixed it here. 👏 Please install the newest version of pg-pool and see if it fixes your issue!

Thanks

If you or your company benefit from node-postgres and have the means, please consider supporting my work on Patreon.