[MCOL-643] support longer then 8m rows in order by Created: 2017-03-28 Updated: 2023-11-17 Resolved: 2023-10-25 |
|
| Status: | Closed |
| Project: | MariaDB ColumnStore |
| Component/s: | Columnstore Select Handler |
| Affects Version/s: | 1.1.0 |
| Fix Version/s: | Icebox |
| Type: | New Feature | Priority: | Minor |
| Reporter: | Andrew Hutchings (Inactive) | Assignee: | Todd Stoffel (Inactive) |
| Resolution: | Won't Fix | Votes: | 3 |
| Labels: | triage | ||
| Issue Links: |
|
||||||||||||||||||||||||
| Epic Link: | ColumnStore Compatibility Improvements | ||||||||||||||||||||||||
| Sprint: | 2017-7, 2017-8, 2017-15, 2017-16, 2017-17, 2017-18 | ||||||||||||||||||||||||
| Description |
|
Without ha_calpont_impl_rnd_pos we are limited to a maximum of 8MB row length including blobs for sorting. For 1.0 this was fine but with BLOB/TEXT support this may not be enough. At the moment this function just returns the following error message: Sorting length exceeded. Session variable max_length_for_sort_data needs to be set higher. |
| Comments |
| Comment by Andrew Hutchings (Inactive) [ 2017-05-11 ] |
|
Parts of what we need are already implemented. sm.cpp has code to save results into band files which can then be read back. We need random read to use SAVING and SAVED states. |
| Comment by Andrew Hutchings (Inactive) [ 2017-11-27 ] |
|
Added related issue In addition we could just use block/offset pointers and request those from PrimProc instead of using a disk cache of the result set. |
| Comment by Andrew Hutchings (Inactive) [ 2020-01-31 ] |
|
There are other reasons to implement this, but the original problem is solved in 1.4 |
| Comment by Valerii Kravchuk [ 2021-08-28 ] |
|
Can we get any status update here? Do we still have any need to implement this feature in 10.6/Columnstore 6.x.y? |