when to use checkpoint vs opening rocksdb in read only mode about rocksdb

[ltamasi]

For example, I have a rocksdb instance that is open in read/write mode all the time and is being written to/read from all the time. I have a separate cronjob (different process) that opens rocksdb in read only mode
...

Questions:

1. in my specific use case, is taking a checkpoint and processing that from a different process better? If so why?

Generally speaking, using a read-only instance this way is not safe/supported: the read/write instance can delete files (e.g. SST files that get compacted away) from underneath the read-only instance. Checkpoints ensure that all the files necessary for a consistent snapshot of the database are preserved (by copying or hard-linking them).

2. I assume the comment about keeping the iterators short lived still applies even when opening rocksdb in read only mode as it will pin the resources in the process that has the db open in read/write mode?

Having a read-only instance open has no effect on the read/write instance, so files are not protected from being deleted (see above).

[zaidoon1]

got it, so aside from the possible consistency issue, is there any performance benefits of using checkpoint over read only instance for my use case?

[ltamasi]

In general, checkpoints do not offer a performance benefit. They don't have much of a performance overhead either though (other than copying certain types of files where hard linking is not appropriate), so I would suggest using them in use cases like this for the sake of correctness.

[zaidoon1]

got it, thank you!