[MDEV-16967] Spider CREATE TABLE Pushdown Created: 2018-08-13 Updated: 2024-01-20 |
|
| Status: | Stalled |
| Project: | MariaDB Server |
| Component/s: | Storage Engine - Spider |
| Fix Version/s: | None |
| Type: | Task | Priority: | Major |
| Reporter: | Jacob Mathew (Inactive) | Assignee: | Unassigned |
| Resolution: | Unresolved | Votes: | 1 |
| Labels: | spider | ||
| Issue Links: |
|
||||||||||||||||||||||||
| Description |
|
Currently it is necessary for a Spidered table to be created separately on each node in the Spider cluster. This process by the end user is time-consuming, repetitive and error-prone. To simplify this, we can have the Spider node push down the CREATE TABLE statement to each of the data nodes. The statement pushed down to the data nodes would not have the COMMENT that is used to specify the connection information. In case of a failure on any of the data nodes, Spider would potentially need to do cleanup to synchronize the table definition across the Spider cluster. Since DDLs are not transactional, the table cleanup would need to be implemented in the Spider node without the benefit of a rollback to remove the table definition throughout the cluster. We would also add a new Spider configuration setting to specify :
|
| Comments |
| Comment by Jacob Mathew (Inactive) [ 2018-08-21 ] |
|
Correct setup of vertically partitioned tables requires an additional level of care and precision over correctly setting up a Spider cluster, and is therefore even more trouble prone. We therefore want to complete the CREATE TABLE pushdown task before proceeding with the tasks pertaining to vertically partitioned tables. |
| Comment by Kentoku Shiba (Inactive) [ 2019-02-22 ] |
|
I just pushed it to bb-10.4-spider-ks. Please review this. |
| Comment by Sergei Golubchik [ 2019-05-19 ] |
|
this commit doesn't change anything outside of storage/spider, so I'm reassigning it back to the original reviewer |
| Comment by Sachin Setiya (Inactive) [ 2019-05-27 ] |
|
Okay to push after a more elaborate commit message. |
| Comment by Federico Razzoli [ 2024-01-20 ] |
|
If in some cases the operation fails and orphaned tables remain on some remote servers, it wouldn't be a big problem. As long as we can re-run a create-or-replace operation. |