Continued work on the WAL

This commit is contained in:
2025-11-17 21:52:44 +01:00
parent 6126a2e932
commit fb28e94a83
8 changed files with 320 additions and 52 deletions
+5
View File
@@ -13,3 +13,8 @@
[ ] Add failt injections to the simulation
[ ] Add authentication
[ ] Add release build and allow switching between debug, release, coverage, sanitizer builds while calling make
[ ] When an operation is committed to the WAL, it may then fail due to not deterministic reasons (out of memory). When the metadata server restarts and replays the log, the operation may succede and cause the system to diverge from the last instance of the process. Is this not a problem? How do we solve it?
[ ] Add notes on who inspired the testing approach
[ ] Add notes on benchmarks and that we don't do batching
[ ] Fix endianess when writing to network and the WAL
[ ] Add a change counter to FileTree. The counter is initialized to 0 and updated any time the tree is changed. When a new file is created, the current change counter is assigned to it. This makes it possible to verify that the file didn't change in between read and write operations without that clumsy previous hash list and previous chunk size.