Paul Dix
Paul Dix
this is done
I've done some initial set of tests and gathered some numbers from my M1 Macbook Pro. The first test was to write the `one_mil` data set into a single segment...
The third set of graphs kept the number of rows constant at 10M and spread them out over varying numbers of files. The number of files didn't seem to impact...
I hacked up a quick and dirty index in the buffer which maps the series ID to the set of rows it appears in. Surprisingly, this didn't have a terrible...
Absolutely, I actually don't use series id in the buffer, but instead just the `usize` pointer to the row. However, I intend to persist the series id in the Parquet...
Using the comparison app in #24889 I ran for five minutes against the non-index buffer and the buffer with an index. You can see the query response times moving up...
This work is largely complete. Optimizations have moved into the file index in Pro.
It's on our todo list to write it up influxdata/docs-v2#6276
The hard delete should just set the option for when to hard delete like in #26262
Users will need some way to track the status and progress of a hard delete as it's something that could take some time. I think the cleanest way is to...