+1 to warn, config, and disable. On Mon, Jan 14, 2019 at 1:45 PM Jonathan Haddad <j...@jonhaddad.com> wrote:
> I'm very much in favor of a warning, and I lean towards disabling them (and > MVs, while we're at it) by default as well. > > I've seen both features be the death of clusters, and are a major risk for > teams that are brand new to Cassandra. > > > > On Mon, Jan 14, 2019 at 11:19 AM Andrés de la Peña < > a.penya.gar...@gmail.com> > wrote: > > > Hello all, > > > > It is my understanding that SASI is still to be considered an > > experimental/beta feature, and they apparently are not being very > actively > > developed. Some higlighted problems in SASI are: > > > > - OOMs during flush, as it is described in CASSANDRA-12662 > > - General secondary index consistency problems described in > CASSANDRA-8272. > > There is a pending-review patch addressing the problem for regular 2i. > > However, the proposed solution is based on indexing tombstones. SASI > > doesn't index tombstones, so it wouldn't be enterely trivial to extend > the > > approach to SASI. > > - Probably insufficient testing. As far as I know, we don't have a single > > dtest for SASI nor tests dealing with large SSTables. > > > > Similarly to what CASSANDRA-13959 did with materialized views, > > CASSANDRA-14866 aims to throw a native protocol warning about SASI > > experimental state, and to add a config property to disable them. Perhaps > > this property could be disabled by default in trunk. This should raise > > awareness about SASI maturity until we let them in a more stable state. > > > > The purpose for this thread is discussing whether we want to add this > > warning, the config property and, more controversially, if we want to set > > SASI as disabled by default in trunk. > > > > WDYT? > > > > > -- > Jon Haddad > http://www.rustyrazorblade.com > twitter: rustyrazorblade >
