Go Bigtable Client Questions

[Derek Perkins]

While implementing the Go Bigtable package, I had a few questions that will affect how I structure it. My initial thought is to maintain a map[tableID]*bigtable.Table and use a single client at the instance level for the life of the instance (see below).

1. Is a Bigtable client / table safe for concurrent access?
2. What's the lifecycle of the client / table? Should I treat it like a MySQL connection and leave the connection open indefinitely?

var (

tables = map[string]*bigtable.Table{}

mu     = &sync.RWMutex{}

client *bigtable.Client

)

[David Symonds]

On 24 July 2015 at 00:29, Derek Perkins <de...@derekperkins.com> wrote:

> Is a Bigtable client / table safe for concurrent access?

Once created, yes.

> What's the lifecycle of the client / table? Should I treat it like a MySQL
> connection and leave the connection open indefinitely?

Yes, you can (and probably should) leave the connection open for the
lifetime of your program if you're going to keep using it. It will
transparently redial when a connection breaks.

[Derek Perkins]

Great, thanks for clarifying that. Is there any issue with a single client being used to access multiple tables? Does every query use its own connection, or is the implementation more like MySQL connection pools so that short queries don't get caught waiting for larger ones?

[David Symonds]

It's probably fine to have a single client for multiple tables. It's
using gRPC and HTTP/2 under the hood, and one long query won't block
other queries from making progress.

[Derek Perkins]

Great, that's super helpful.