snapshots: disable peer selection when no snapshot sources are selected #6594
Conversation
cali-jumptrading
force-pushed
the
cali/snapshot-tiles-http-resolver-2
branch
2 times, most recently
from
7c47c6f to
972fa5d
Compare
There was a problem hiding this comment.
I'm not convinced we need this config option. It seems like we can just use the combination of
- entrypoints disabled
- all gossip peers disabled
- http servers empty/disabled
to mean the same thing. We should support this in production anyway (the operator is saying they will provide their own snapshot on disk)
cali-jumptrading
force-pushed
the
cali/snapshot-tiles-http-resolver-2
branch
4 times, most recently
from
f58c238 to
0420e31
Compare
cali-jumptrading
changed the title
cali-jumptrading
force-pushed
the
cali/snapshot-tiles-http-resolver-2
branch
3 times, most recently
from
223405d to
0c3c629
Compare
cali-jumptrading
force-pushed
the
cali/snapshot-tiles-http-resolver-2
branch
from
0c3c629 to
28a5ecc
Compare
| FD_LOG_ERR(( "Local snapshot `%s` is too old and downloading new snapshots is disabled. " | ||
| "Please enable downloading via [snapshots.download] and restart.", ctx->local_in.full_snapshot_path ) ); |
There was a problem hiding this comment.
Do we have any existing precedent for LOG_ERR'ing here? This is a situation where the validator isn't expecting to make progress / start up correctly, but it's not really a "crashable" edge case...
So is there some precedent for "alert loudly and keep running" or do we prefer to crash? Presumably most operators have their validators running under systemd or whatnot which will auto restart if it crashes/stops, which won't give them any chance to correct the config
There was a problem hiding this comment.
Separately not sure how I feel about asking the operator to enable downloading. Maybe they just need to get a newer snapshot some other way external to our code. We should also probably log the local slot, cluster slot, and the max slot age so they can see those things
There was a problem hiding this comment.
Agree with logging the local slot, cluster slot, and the max slot age. Not sure if we have precedent for alert loudly and keep going, or crashing. We added this download config option because agave had something similar: no_snapshot_fetch, which disabled downloading snapshots and only loaded from local snapshots if present.
No description provided.