Cassandra handle load well at scale. > Power outage with fsync mode and for example 5 seconds window guarantee that >after down only 5 seconds is lost?
I'd recommend a set up that preserve invariant regarding database consistency. I'd recommend not using commitlog_sync: periodic in a mono node setup. Cf earlier mails.-- Best regards, Benoit TELLIER General manager of Linagora VIETNAM. Product owner for Twake-Mail product. Chairman of the Apache James project. Mail: [email protected] Tel: (0033) 6 77 26 04 58 (WhatsApp, Signal) Le sept. 30, 2026 1:01 PM, de Ilya Terskov <[email protected]>Cassandra under load pretty stable? Power outage with fsync mode and for example 5 seconds window guarantee that after down only 5 seconds is lost? ср, 30 сент. 2026 г., 17:57 Benoit TELLIER via server-user <[email protected]>: We officially tested only cassandra -- Best regards, Benoit TELLIER General manager of Linagora VIETNAM. Product owner for Twake-Mail product. Chairman of the Apache James project. Mail: [email protected] Tel: (0033) 6 77 26 04 58 (WhatsApp, Signal) Le sept. 30, 2026 9:20 AM, de Ilya Terskov <[email protected]>and in the end which CQL DB in this line is best suited for standalone, most stable (power outage?) and 100% compatible with James? hah yeah docs.yugabyte.com/stable/explore/ycql-language/ "YugabyteDB's YCQL API has its roots in the Cassandra Query Language (CQL)" Please i BEG you to test it and tell me its 100% then i prefer it! ср, 30 сент. 2026 г. в 14:07, Benoit TELLIER via server-user < [email protected]>: > ScyllaDB has same architecture than Cassandra. > > Let me correct slightly my previous sayings... > > Cassandra commit log can be configured to be synchronous (commitlog_sync: > group) in which case commitlog_sync_group_window allow to configure how > long the write waits to trigger another (grouped) flush eg 20ms. > This makes it viable in a mono-node setup at the cost of a bit of write > throughtput. > > Tombstones are for multi-node setup, but yes being myself a Cassandra > nerd, Cassandra architecture is very cool!We ran distributed app on systems > with hundred-thousand users and million-per-day messages, so yes it is > mature. The intent with the PG backend is to make it easier and cheaper to > operate to make self hosting a viable option. > > -- > > > Best regards, > > Benoit TELLIER > > General manager of Linagora VIETNAM. > Product owner for Twake-Mail product. > Chairman of the Apache James project. > > Mail: [email protected] > Tel: (0033) 6 77 26 04 58 (WhatsApp, Signal) > > > > Le sept. 30, 2026 8:45 AM, de Ilya Terskov <[email protected]>Well > thats bad, there no DB with CQL compliance (like ScyllaDB) but for > standalone? :c i like idea of Tombstones even sounds cool and mechanics too > and as i understand disturbed-app better tested? > > ср, 30 сент. 2026 г. в 12:06, Benoit TELLIER via server-user < > [email protected]>: > > > Answer: very bad > > The point is that Cassandra is a distributed system, and relies on it for > > data safety. > > Eg the commit log is fsynced periodically, and reliability is achieved by > > replicating to other nodes. In mono node setup, this 'eans a fault leads > to > > data loss. > > Cassandra make big compromise which allow eg running 8 James on top of 20 > > Cassandra for a critical system, but it comes at a price: harder > > programming model and eventual consistency semantics.... > > -- > > > > > > Best regards, > > > > Benoit TELLIER > > > > General manager of Linagora VIETNAM. > > Product owner for Twake-Mail product. > > Chairman of the Apache James project. > > > > Mail: [email protected] > > Tel: (0033) 6 77 26 04 58 (WhatsApp, Signal) > > > > > > > > Le sept. 30, 2026 1:30 AM, de Ilya Terskov <[email protected]>Good > > morning guys! > > > > Learning about Cassandra make me really wanna try use it, how it good > with > > standalone single server? I see in 5.x and 6.x versions it became much > more > > stable and got acid right? Compared to postgres its same level of > stability > > in terms of power outrage? > > > > >
