This paper that improves clock synchronization accuracy to tens of nanoseconds is amazing work. What changes with that level of accuracy? Like the paper says, Spanner commit times get a lot shorter. But is it now safe to take action without coordination? https://www.usenix.org/system/files/conference/nsdi18/nsdi18-geng.pdf …
-
-
And IO. To use this pattern you need to be super sure that IO like logging and metrics don't add hidden pauses. And GC too. As you say, lots of ways to get this wrong in practice.
-
A common issue we've seen it doing logging to the OS page cache (blocking, but not sync-to-disk). Usually super fast, but can take a really long time under other IO pressure. Meanwhile, leader thinks the clock isn't moving.
- Još 2 druga odgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.