What happens to a Rayls payment if validators crash mid-transfer?
I built Axyl, the Rayls consensus client, from source, ran 4 validators on my laptop and crashed them during live payments.
One validator down: 240 of 240 payments settled. Longest pause 2.58 s.
Two validators down: the chain stopped until they came back. A node still accepted my payment in 15 ms with no error, and it settled 41 s later. All 45 payments sent during those outages settled, 8 to 69 s after submission.
Same nonce sent to two validators, 24 times: both returned a hash, but only one ever settled. No double spend.
So on this setup it stalls and does not fork. The lesson for apps: a returned hash does not mean paid.
On the public testnet, 150 payments had a median of 2.44 s from submit to receipt, as seen from one client. That includes about 0.4 s of network round trip.
Limits: one laptop, crashes only, small samples, no mainnet payments. One restarted validator diverged in an early messy run. I could not reproduce it and reported it privately to Rayls.
Data and code: https://github.com/anshxft/rayls-axyl-failure-tests
Not financial advice.+
I built Axyl, the Rayls consensus client, from source, ran 4 validators on my laptop and crashed them during live payments.
One validator down: 240 of 240 payments settled. Longest pause 2.58 s.
Two validators down: the chain stopped until they came back. A node still accepted my payment in 15 ms with no error, and it settled 41 s later. All 45 payments sent during those outages settled, 8 to 69 s after submission.
Same nonce sent to two validators, 24 times: both returned a hash, but only one ever settled. No double spend.
So on this setup it stalls and does not fork. The lesson for apps: a returned hash does not mean paid.
On the public testnet, 150 payments had a median of 2.44 s from submit to receipt, as seen from one client. That includes about 0.4 s of network round trip.
Limits: one laptop, crashes only, small samples, no mainnet payments. One restarted validator diverged in an early messy run. I could not reproduce it and reported it privately to Rayls.
Data and code: https://github.com/anshxft/rayls-axyl-failure-tests
Not financial advice.+