Improve FloodfillVerifyStoreJob success rate
Only about 50% from looking at netdb.floodfillVerifyXXX stats. Seems like it should be much higher, but I don't have historical data. Tested on 2.2.1 and 2.3.0-11 with similar results; unlikely to be related to subdb changes. The high failure rate leads to increased netdb traffic, bandwidth, and load on floodfills.
Possible causes:
- Poor selection of store peer
- Poor selection of verify peer
- Reusing verify peer that failed previously
- Large number of poor/malicious floodfills we aren't avoiding
- Different view of closest floodfills from the store peer, due to FloodfillPeerSelector heuristics, i2pd differences, or other
- Bugs
- Stat collection issues
- Poor reliability of expl. tunnels
- Conn limits at ffs, OBEP, or IBGW
- Encryption compatibility
- i2pd bugs
- Should we use congestion caps as part of ff selection?
To be researched.