Timeout on read does not work?

[alexey.rodri...@soundcloud.com]

Hi there!

I have a question about controlling the timeout behaviour of Cassandra queries.

I am trying to set up a Cassandra client using gocql. This client is very simple in that it only performs look ups and it doesn't update the database. I want my client to timeout when Cassandra does not respond within a given time frame. I noticed that there is the Timeout parameter in ClusterConfig, and that NewCluster (which I use) sets it to 600ms. However, I am not getting read errors when this timeout is exceeded, I have set this timeout very low (1ms for example) to check whether this parameter works, but I still don't get a timeout error.

I am missing something on how this parameter should be used?

Thank you!

Alexey

[Ben Hood]

Hi Alexey,

Could you post a short running example of the code that you are having
issues with please?

Cheers,

Ben

> --
> You received this message because you are subscribed to the Google Groups "gocql" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to gocql+un...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[Alexey Rodriguez Yakushev]

Hi Ben, here is a copy paste of the offending code with slight edits to remove irrelevant details:

cluster := gocql.NewCluster(hosts...)

cluster.Keyspace = ...

cluster.Consistency = gocql.Quorum

cluster.Authenticator = gocql.PasswordAuthenticator{...}

cluster.Timeout = 50 * time.Millisecond // Low deadline for testing purposes

session, err := cluster.CreateSession()

if err != nil {

log.Fatalf("could not connect to cassandra cluster %s", err)

}

return &cassandra{

session: session,

}

And later in the code the session is used (s is a cassandra struct that contains the session field):

start := time.Now()

if err := s.session.Query(query).Scan(&text); err != nil {

return ..., fmt.Errorf("could not select: %s", err.Error())

}

elapsed := time.Since(start)

fmt.Printf("Response time: %v\n", elapsed)

A printout of a run:

connect: failed to connect to "IP:9042": dial tcp IP:9042: i/o timeout

connect: failed to connect to "IP:9042": dial tcp IP:9042: i/o timeout

Response time: 137.591482ms

Interestingly, it seems that the latency checking is performed when the connection is performed (if I decrease Timeout, the connection code will abort after a few timeout errors). Unfortunately, the latency checking is not performed for the query itself.

Cheers,

Alexey

[Alexey Rodriguez Yakushev]

In case the query is relevant (I wouldn't think so), here is how it looks like:

const (

Query = "SELECT value FROM %s WHERE key = %s"

)

On Mon, Jun 15, 2015 at 7:16 PM, Ben Hood <0x6e...@gmail.com> wrote:

[Chris Bannister]

Currently there is a limitation with the query sending code is that it will only set read timeouts once it has received a header because otherwise the conn read loop can be looping for a long time without any data on the socket.

I'll create an issue and start working on it, its quite a simple fix.

[Alexey Rodriguez Yakushev]

That's great Chris.

I'll be tracking this issue then:

https://github.com/gocql/gocql/issues/416

Cheers,

Alexey