non symmetric sync? #2319
Comments
|
so far what I have found to be a notable difference between successful runs and this one is that the expected entries are received all via gossip in successful runs. For the silent node, there is no gossip message and no We should look into why gossip propagation is failing. However, gossip should not be sole provider of info if sync has already started between both nodes. Basically, this reinforces my hypothesis that it's sync itself which is not occurring in a symmetric manner. I'll give it a go to reproduce the failure by disabling/breaking gossip |
|
from testing by ignoring entries that come from gossip, nothing seems to work at all. I would expect set reconciliation to be enough to obtain the metadata of the missing entries. If this is not the case then what's set reconciliation providing that gossip is not supplying already? |
These are the full logs of a test failure for
test_download_policies. This is not a frequent failure at all so having the logs is a great opportunity to understand what's going on with this sync/docs problem.The test setup is as follows:
nkpqjtgrfsdzojzk(NK) andwp3g26raw2ercopv(WP),NKhas 6 locally inserted entries.WPhas 3 locally inserted entries.NKshould download 1 of the entriesWPhas.WPshould download 3 of the entriesNKhas.What actually happens:
WPemits theRemoteInsertevents it shouldWPdownloads the entries it shouldNK, noRemoteInsertat all, and of course no downloads either.Notable logs:
NKclaims to be starting sync with 0 peers. This is weird and potentially a problem on itself. However the fact thatWPis actively communicating and successfully syncing withNKtells me there is a connection and sync messages are being passed. Still,NKis not using that info on its side, which leads me to conclude sync is not happening symmetricallyThe text was updated successfully, but these errors were encountered: