[jira] [Comment Edited] (IGNITE-14556) Live Schema. Add Tuple validation.

["Andrey Mashenkov (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/IGNITE-14556?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17322239#comment-17322239 ] 

Andrey Mashenkov edited comment on IGNITE-14556 at 7/6/21, 9:53 AM:
--------------------------------------------------------------------

Do we need an option to relax these checks for STRICT mode? and allow arbitrary fields in the Tuple, but ignore non-relevant while building a Row?

Maybe this could be a default option? If so, we only need this validation for LIVE-schema to detect schema upgrades.

 


was (Author: amashenkov):
Do we need an option to relax these checks for STRICT mode? and allow to pass any fields within the Tuple, but ignore non-relevant while building a Row?

Maybe this could be a default option? If so, we only need this validation for LIVE-schema to detect schema upgrade.

 

> Live Schema. Add Tuple validation.
> ----------------------------------
>
>                 Key: IGNITE-14556
>                 URL: https://issues.apache.org/jira/browse/IGNITE-14556
>             Project: Ignite
>          Issue Type: Improvement
>            Reporter: Andrey Mashenkov
>            Priority: Major
>              Labels: iep-54, ignite-3
>             Fix For: 3.0.0-alpha3
>
>   Original Estimate: 48h
>  Remaining Estimate: 48h
>
> h3. Motivation.
> At a point of Table public method is called by a user, we need to validate user input (for LIVE-schema purposes at least).
> h3. Description.
> We can add this logic to check if value fields match the current schema version (no new fields).
>  * For LIVE-Schema. If Tuple has one or more additional columns, then we should try to register a new schema first, then proceed with the user operation.
>  * For STRICT-Schema.  If Tuple has one or more additional columns, then we should fail the user operation.
>  * For KeyValueView, we should validate key Tuple as well, and fail if there are unknown columns. Because a key column span is immutable. The only exception may be if a user creates a schemaless table, then a schema of the 1-st version should be registered instantly.
> Assumed, any column type mismatch or missed Non-Nullable columns will be caught and processed by RowAssembler.
> h4. Possible optimization.
> It is possible to add the validation into a TupleBuilder and then just check the Tuple instance class (should be a builder). 
>  For any Tuple of unknown type or if a schema was changed concurrently (TupleBuilder validated input against outdated schema version), then fallback to default logic and re-validate input against the latest schema.
>  



--
This message was sent by Atlassian Jira
(v8.3.4#803005)