Tyler Crain

This is loaded from the log: https://github.com/SyncFree/antidote/blob/master/src/logging_vnode.erl#L269 It is done during "normal startup or node restart": https://github.com/SyncFree/antidote/blob/master/src/inter_dc_manager.erl#L126 It should be loaded on process restart, a suggestion for fixing this is...

For information I did some tests on antidote to see the impact of the materialiser, all experiments are done using counter CRDT types, 2000 keys, single operation transactions either performing...

I guess at this point I haven't investigated closely enough to tell exactly what is taking the most time, so the eagerness/laziness of generating snapshot it definitely something we need...

Not a fix, but a suggestion is to run NTP on all nodes in a DC before any of the erlang VMs are started. My understanding is that "time warp"...

I forgot his one: clock_si_vnode : This process keeps in memory the transactions that have been prepared, but not yet committed. On restart it should load this information from the...

While the Antidote protocol itself is designed specifically to work under the presence of faults, recovering from these is mostly unimplemented (mostly just because of the lack of people with...

Another simple optimisation that is missing is to allow single key update transactions to not require using the 2 phase commit protocol.

Ok, it looks like I introduced this when add the recovery from the log, so I will fix it. The test you mention will fail on master? Can I use...

Can you also check if you need a check here for the journal tool, otherwise looks good, thanks: https://github.com/Alluxio/alluxio/blob/master/core/server/master/src/main/java/alluxio/master/journal/tool/AbstractJournalDumper.java#L88

> * Theoretically writes were supposed to inherit the parent folder's (bucket's) mode bits of `770` but in reality they were committed as `755` for directories and `644` for files....