vtgate engine: command line flag to generate sequences for 0 columns #4751
Conversation
This diff introduces -seq-auto-value-on-zero as a vtgate flag. Until now sequences have only populated NULL columns, which is not consistent with the default behavior of mysql. It's consistent with the NO_AUTO_VALUE_ON_ZERO sql mode. In a perfect world we might want to detect per-keyspace sql modes. For now, it seems simple and adequate to let people select at the vtgate level if they want this behavior or not for all keyspace. The motivation for this diff is that we have some situations where people still test code against direct mysql without vtgate. We've twice had folks commit insert statements with 0 for a sequence column that worked in a local test but failed against vtgate. Aligning behavior of vtgate with mysql seems like the cleanest way to prevent that in the future. Signed-off-by: David Weitzman <dweitzman@pinterest.com>
b16bd67 to
7673b28
Compare
I kinda think the best thing to do is make this the default and only behavior rather than supporting both. I'm having trouble imagining a real scenario where someone would want to enable NO_AUTO_VALUE_ON_ZERO. Seems like if push comes to shove in some kind of odd emergency where it's super important to insert a row with ID 0 you could always temporarily disable sequences or have a dba bypass vitess or for the moment you could use v1 or v2 destinations to bypass sequences. If both modes were supported by vtgate, I don't see much value in being able to switch between them dynamically. Even if switching were supported, I wouldn't want to require all db connections to explicitly and individually opt-in to the normal default mysql behavior. |
|
If this is the mysql default, I'm fine with changing behavior without a flag. |
|
@dweitzman ping |
|
We're interested in this and @MordFustang21 could take it over if @dweitzman doesn't have time. @sougou Do you feel differently about this now? Do you want the |
Feel free to take it over. It's a small diff. (One reason I never finished this was that I was feeling conflicted about how session variables should be handled) |
Signed-off-by: David Weitzman <dweitzman@pinterest.com>
|
@harshit-gangal @systay looks like this can be closed now? |
|
Closing in favour of #6749 |
This diff introduces -seq-auto-value-on-zero as a vtgate flag.
Until now sequences have only populated NULL columns, which is not consistent with
the default behavior of mysql. It's consistent with the NO_AUTO_VALUE_ON_ZERO sql mode.
In a perfect world we might want to detect per-keyspace sql modes. For now, it seems
simple and adequate to let people select at the vtgate level if they want this
behavior or not for all keyspace.
The motivation for this diff is that we have some situations where people still test code
against direct mysql without vtgate. We've twice had folks commit insert statements
with 0 for a sequence column that worked in a local test but failed against vtgate.
Aligning behavior of vtgate with mysql seems like the cleanest way to prevent that
in the future.