I think there is a bug in safe vault code when several nodes disappear from a section at the same time.
I have created a small test network with min section size = 4 in routing config file. My network has normally 4 nodes which means that all data chunks are duplicated 4 times with one copy in each vault. I have developed a utility to display the number of immutable data chunks and the number of mutable data chunks in each vault. And I observe that these 2 numbers are the always the same in all the vaults.
This kind of setup is useful to observe the number of chunks created by a command: for example this is the way I found this issue.
At one time I deleted 2 vaults by accident in a short interval (I didn’t notice the delay between them). No data was lost because the 2 remaining vaults had still the same values for these numbers. Then I relaunched 2 vaults (one by one, with an enough delay between them) but the 2 new vaults display lower values for these numbers, meaning that some data is not stored in these vaults.
This means that some chunks are duplicated less than 4 times which is not a normal state. This fundamental invariant is not respected, which can be the source of ulterior loss of data.
I think this problem is possibly the root cause of the bad fate of the first alpha 2 network and other past networks. I was considering launching a community network (and this test network was in preparation of it) but the discovery of this problem is blocking for that.
Could someone at @maidsafe analyze this problem?