Ayende Rahien

The issue is that we `fsync` there before hand. So there is inconsistency between the files.

ARM64 shouldn't matter. Can you print the `/Databases/CUSTOMERNAME1/` files including modification dates?

``` 7 -rw------- 1 dhcpcd dhcpcd 162 Mar 25 17:46 headers.one 7 -rw------- 1 dhcpcd dhcpcd 162 Nov 29 05:30 headers.two ``` This is *very* suspicious. `headers.two` behind from Nov?...

What do you mean synced, the last time you wrote to the database? If that is the case, Nov make sense then Here is the problem, what we see is...

Can you send the `headers.one` file? (Not user data is located there, FYI)

Okay, this is _weird_. I looked into this, and it doesn't make sense. The first part of a `header` file is a magic number, which should be: `0xB16BAADC0DEF0015` In the...

Can you send _both_ of those files as binary (zip, etc)? I think that roundtripping through `hexdump` messed things up, yes.

Thanks, that was very helpful. The files are now workable. Looking into this, `headers.one` is the newer version, and the last checkpoint is from journal `9`, transaction `191`. Can you...

Thanks, that was really interesting. Here is my finding, in journal 9, we see transaction commits from 158 .. 191. Representing byte ranges of 0 ... 12,967,936. Seems valid so...

If you can replicate, that would be amazing yes. I'm not aware of anything in ZFS that would cause this, but possible. I'm leaning more toward the k8s setup