5 ms·
The common trend among the outages(atleast Gmail, Amazon EC2) seems to be that the queuing of the transaction messages seems to be overwhelming the network thus
by recoiledsnake 15y ago
The common trend among the outages(atleast Gmail, Amazon EC2) seems to be that the queuing of the transaction messages seems to be overwhelming the network thus further making the attempts at recovery futile.
Maybe a solution is to immediately stop all new requests and have servers that record all the transactions to disk instead of redirecting them to overloaded servers.
- acqq 15y agoHow can you make people stop sending messages? How can "recording to disks" make some change? What do you think is being done otherwise than writing some databases?
- gwright 15y agoYou have to build this sort of think into the infrastructure. For example, TCP and Ethernet retransmission strategies and HTTP servers should have reasonable timeouts that result in error responses rather than just letting incomplete HTTP sessions pile up. The same principles can and should be applied to any communication protocol.