Question: predis multi slaves with a slave down

[endymon]

My question is, with a predis setup involving multiple configured slaves, If 1 or more slaves are down, how do you get predis to stop trying to send queries to the down slave(s)?

When I tested taking a slave down I got spammed by these:

Predis\\Connection\\ConnectionException: from 'EXISTS XXXXXX{THE_KEY}XXXXX' while connected to host: 'XXX{THE_HOST}XXX' MESSAGE: Connection refused [tcp://localhost:XX{PORT}XX]

And then after a short while the error message switched to:

'LOADING Redis is loading the dataset in memory'

These exceptions make sense if there is only one slave, but I would have assumed that in a multiple slave configuration predis either would by default (or by configuration parameter) fail-over silently to a NON down slave host. ONLY if all slaves were down could I see an exception propagating up to the client, but not in this case.

How does predis handle multiple slaves when 1 or more are failed or unavailable?
Is there such a configuration parameter?

Thanks
-- Matt

[xiconfjs]

Right now we are testing the replication-sentinel function of the predis 1.1-sentinel branch. And we are experiencing the same phenomenon. Is there an option to skip a slave which is re-syncing itself?

[nrk]

@endymon starting with Predis v1.1, trying to execute a read-only command on a unreachable slave will discard that slave and try again on the next one, or even the master if all of the slaves are offline (see the related changes). As for errors such as -LOADING they are not handled yet but I guess it should be fairly easy to intercept them to apply the same logic described above.

@xiconfjs the same applies to redis-sentinel, I will look into it.

[nrk]

Now Predis intercepts -LOADING responses, removes the "busy" slave from the pool and executes the command on the next slave (or master as last resort) instead of throwing an exception right away.