Agata Kruszona-Zawadzka

There are two separate issues raised in this thread: - compression of chunk data on chunkserver - this is on our roadmap, but it doesn't have a high priority, as...

Did you check logs when this happened or can you still check them? Look for a message that the client lost connection to master and reconnected. If it is there,...

It's on our roadmap to display some summary of error/warning logs in CGI, this would include network errors too.

Were they shown on the chart as **successful** replications? Can you post a screenshot?

By saying "indistinguishable from normal (successful ones)" - you are saying the spike on your chart was all in one color, right?

It's possible, that the algorithm that tries not to kill a brand new hard drive is misbehaving in this case. We will look into it.

We cannot do it on-line, this would take too much resources. We could make statistics with every chunk loop completed by the master and store information for each storage class:...

Ok, I'm adding it to the feature list then.

Are there snapshots in your MooseFS instance?

This can happen when you use snapshots. When we introduced labels in MooseFS, we had to do something with snapshots. Scenario: - you record files with, let's say, 4 million...