No possible successful shutdown for --fast-and-loose

[tsutsu]

I'm running:

• Which Parity version?: Parity/v1.10.0-unstable-f8bf7e7d4-20180118/x86_64-macos/rustc1.23.0
• Which operating system?: MacOS
• How installed?: from source
• Are you fully synchronized?: no
• Which network are you connected to?: ropsten
• Did you try to restart the node?: yes (that's the problem!)

Quoting parity --help (emphasis mine):

--fast-and-loose
Disables DB WAL, which gives a significant speed up but means an unclean exit is unrecoverable.

Expected behaviour: that there is actually such a thing as a "clean" exit, that won't corrupt my DB when using --fast-and-loose.

Actual behaviour:

• the database is corrupted when you press Ctrl+C.
• the database is corrupted when you send killall -HUP parity.
• the database is corrupted even if I ask very nicely :(

Looking at the source, there's no cleanup-DB-on-quit code in either parity/main.rs or parity/run.rs. And, in fact, ethcore/src/blockchain/blockchain.rs never even calls anything like KeyValueDB::close().

[Aggour]

The --fast-and-loose argument doesn't allow for a clean exit on purpose. It's a feature meant to be used for quick synchronization on testnet and not for use on the production network.

[andresilva]

The reason you don't see any call like KeyValueDB::close() is because the database is closed automatically when it is dropped, that's done here. We fixed an issue recently where some values weren't being dropped correctly on shutdown (#7695) and I think your build doesn't include that. I just ran with --fast-and-loose and exited a couple of times with ctrl-c and there was no db corruption. Try a recent nightly or build the latest master and check if it still happens.

[andresilva]

Okay I just corrupted it, anyway don't use --fast-and-loose. 🙈