"Unable to determine current in source directory" always

samd · **Joined:** Fri Oct 05, 2007 3:22 am **Posts:** 69

I continuously encounter this no mater what I do running Hibernate Search in a clustered configuration.
There is no getting rid of it which messes up search.

Is anyone else using Hibernate Search in a clustered configuration based on the recommended approach? If so do you encounter this?

I think the default shipped slavedirectoryprovider has some potential issues with how they deal with markers?

Code:

private File determineCurrentSourceFile() {
            File sourceFile = null;
            if (new File(source, "current1").exists()) {
                sourceFile = new File(source, "1");
            } else if (new File(source, "current2").exists()) {
                sourceFile = new File(source, "2");
            }
            return sourceFile;
        }

Why are FileLocks not used? Here we are just hoping we don't encounter a race condition which is entirely possible.

Now, on a rare occasion I may see "Unable to determine current in source directory" but always?

The error is occurring in a production environment and this server is not directly accessible making it difficult to debug.

sanne.grinovero · **Posted:** Thu Nov 26, 2009 7:09 pm

Quote:

Is anyone else using Hibernate Search in a clustered configuration based on the recommended approach?

Yes, twice installed myself on moderate load, and heard of several big deployments.

Quote:

If so do you encounter this?

Never seen before.. might have been a different version?

Quote:

Why are FileLocks not used? Here we are just hoping we don't encounter a race condition which is entirely possible.

right it's not looking very well, but it's impossible in practice for a race condition to happen as the switching of current directory is single threaded and is the same thread which is performing this check.

Sounds more like there's some mistake in configuration? It's quite confusing about which one is the target, look out for which configuration is relevant for the master and which one for the slave.
Especially because of the single-threaded design mentioned above, if you're seeing race conditions it's likely the other node is messing a Directory it shouldn't touch.

amin-mc · **Joined:** Wed Oct 03, 2007 2:31 pm **Posts:** 205

Hi

We did a whole bunch of load tests using master slave configuration and we did not experience any issues. I remember when i started on Hibernate search i came across this problem ( i can't remember what i did to get it to work ..not much help) but it was something related to my configuration. Is it possible to get a snippet of your slave config and master config?

It may be worth some updates to the online documentation in relation to the file setup as the Hibernate Search in Action covers this very well.

Haven't helped much but it's worth definitely considering Sanne's last statement.