This post is all about the failings of version 6.3 but it buries the lede:
Version 6.4 includes fixes for all the issues discussed here: stale reads, dirty reads, lost updates (due to both partition detection races and invalid recovery plans), and read-only transaction reordering are all fixed, plus several incidental bugs the VoltDB team identified. After 6.4, VoltDB plans to introduce per-session and per-request isolation levels for users who prefer weaker isolation guarantees in exchange for improved latency. VoltDB’s pre-6.4 development builds have now passed all the original Jepsen tests, as well as more aggressive elaborations on their themes. Version 6.4 appears to provide strong serializability: the strongest safety invariant of any system we’ve tested thus far. This is not a guarantee of correctness: Jepsen can only demonstrate faults, not their absence. However, I am confident that the scenarios we identified in these tests have been resolved. VoltDB has also expanded their internal test suite to replicate Jepsen’s findings, which should help prevent regressions.
I read the top part and thought 'oh, another system that fails to meet their claims' but it's pretty impressive that they did the work to fix it. Nice job.
Version 6.4 includes fixes for all the issues discussed here: stale reads, dirty reads, lost updates (due to both partition detection races and invalid recovery plans), and read-only transaction reordering are all fixed, plus several incidental bugs the VoltDB team identified. After 6.4, VoltDB plans to introduce per-session and per-request isolation levels for users who prefer weaker isolation guarantees in exchange for improved latency. VoltDB’s pre-6.4 development builds have now passed all the original Jepsen tests, as well as more aggressive elaborations on their themes. Version 6.4 appears to provide strong serializability: the strongest safety invariant of any system we’ve tested thus far. This is not a guarantee of correctness: Jepsen can only demonstrate faults, not their absence. However, I am confident that the scenarios we identified in these tests have been resolved. VoltDB has also expanded their internal test suite to replicate Jepsen’s findings, which should help prevent regressions.
I read the top part and thought 'oh, another system that fails to meet their claims' but it's pretty impressive that they did the work to fix it. Nice job.