Create-snapshot check if slot is available on startup

[apfitzge]

Problem

create-snapshot will load bank_forks from a snapshot and try to replay up to snapshot_slot. If snapshot_slot does not exist, it will replay everything available and then fail. This can be slow.

Summary of Changes

Add a check to see if snapshot_slot is available in the blockstore BEFORE trying to load the bank_forks. If it's not we exit since we won't be able to create a snapshot.

Preliminary PR for minimized snapshot mode.

Fixes #

[codecov]

Codecov Report

Merging #25329 (bd00d86) into master (69a0ff9) will increase coverage by 0.0%.
The diff coverage is 66.9%.

@@            Coverage Diff            @@
##           master   #25329     +/-   ##
=========================================
  Coverage    82.0%    82.1%             
=========================================
  Files         598      614     +16     
  Lines      165882   169323   +3441     
=========================================
+ Hits       136125   139029   +2904     
- Misses      29757    30294    +537     

[steviez]

I like the thinking behind adding this check; however, I think we should go even further and check to make sure we have full slots all the way up to (and including) snapshot_slot. It may not be super common to have "gaps" on a validator (although there are instances where we purge slots), a common case is copying a ledger contents (shreds + snapshot) off of validator onto some other machine to debug. Checking the full range would be good to prevent human error in this scenario.

This would definitely add some overhead, but my gut says that in relation to the amount of time it takes to verify and create the snapshot, this extra overhead should be relatively small

[steviez]

In addition to confirming we have a meta, we should make sure that the slot is also full with SlotMeta::is_full()

[apfitzge]

I think we should go even further and check to make sure we have full slots all the way up to (and including) snapshot_slot.

I think that's reasonable. We'd need to check the slots in the range old_snapshot_slot..=new_snapshot_slot, where old_snapshot_slot is the slot of the snapshot we loaded from. Is that right?

And if we are using the --no-snapshot option, it'd be from 0..=new_snapshot_slot.

[steviez]

I think that's reasonable. We'd need to check the slots in the range old_snapshot_slot..=new_snapshot_slot, where old_snapshot_slot is the slot of the snapshot we loaded from. Is that right?

Correct; I think there is some logic in load_bank_forks that you could either replicate or break into a sub-function to get highest available snapshot slot.

And if we are using the --no-snapshot option, it'd be from 0..=new_snapshot_slot.

Correct

[apfitzge]

While testing this change, I found a bug w/ the --no-snapshot option in ledger-tool due to the recent addition of --snapshot-archive-format. Opened a PR here: #25368

[steviez]

LGTM assuming you've done some testing (including cases where slots was not full and/or missing)

Pull request has been modified.