For 1, It’s more about getting server startup faster. Otherwise it may take long time to do rolling restart.
For 2, we can keep adding per column basis index to existing segment, just need to support dictionary and non-dictionary conversion on the fly. Meanwhile we can keep track of real size for used index size as well as segment total size to decide when to do compaction. On Sat, Feb 15, 2020 at 11:35 Mayank Shrivastava <mayanks.apa...@gmail.com> wrote: > Good thought, but need more clarity. For cases that warrant reload > anyways, how does faster server startup help? > For 2, I see this as a precursor to adaptive indexing on the fly. > > Sent from my iPhone > > > On Feb 14, 2020, at 7:55 PM, Fu Xiang <fx19880...@gmail.com> wrote: > > > > 1. For large table, the segment reloading may take very long time during > > server start up. Shall we consider load segment as it is during server > > startup, then reload each segment afterward? > > 2 .Shall we consider when to regenerate the segment. E,g. indexing > removal, > > update dictionary type from dict to non-dict (not yet supported). > > > > -- > > Xiang Fu > > --------------------------------------------------------------------- > To unsubscribe, e-mail: dev-unsubscr...@pinot.apache.org > For additional commands, e-mail: dev-h...@pinot.apache.org > > -- Xiang Fu
