[
https://issues.apache.org/jira/browse/SOLR-15150?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17283283#comment-17283283
]
Chris M. Hostetter commented on SOLR-15150:
-------------------------------------------
David: yeah, we probably intermix "atomic" and "partial" too much, and conflate
the atomic nature of the partial updates with the atomic nature of optimistic
concurrency updates – even when the users aren't doing that.
So agreed: better to use "partial" here to clarify what aspect we're dealing
with.
Ishan: I had originally considering the verb "force" (and i realize now it stil
lingers in a test variable) but it felt really missleading and I'd prefer we
avoid it...
Through the lens of a novice user "forceX" makes me think I'm telling solr "Hey
solr do X even if you wouldn't by default and even if it might break something"
... similar to {{"rm -f"}} or {{"git push --force"}} ... but what we want to
convey is *NOT* that this is a way to _force_ the update to be done in-place
(because we can't actually promise that) .. what we want to convey is that this
is way for the user to say "When i do a partial update, I expect that _either_
the update be done in place, or it must fail"
"Require" felt a little better, because it seemed more like a _request_ if that
makes sense? ... "I require that Solr do X" felt like a good english language
equivalent, to the sentiment I was going for, because Solr can either satisfy
the request, or so "I'm not capable of doing that" (ie:fail)
(I had briefly considered using the verb "assert" as in
{{"assert.inplace.atomic.update"}} or {{"update.partial.assertInPlace"}} but
that felt too Java/C-ish for general solr users)
Omitting any "verb" (either "force" or "require") and just go with something
like {{"update.partial.inplace"}} is an interesting idea, but i feel like it's
too ... "weak" is the closest word i can think of i guess? ... compared to how
this would behave: {{update.partial.inplace=true}} it feels more like a way to
override a default, (that might otherwise be hueristicly determined) and makes
me think that {{update.partial.inplace=false}} would be a way to indicate that
solr should _never_ do my update inplace, which isn't want we want the
(default) {{"false"}} value to mean.
----
So I think on balance at this point I like David's suggestion of
{{"update.partial.requireInPlace"}} the best so far?
> add request level option to fail an atomic update if it can't be done
> 'in-place'
> --------------------------------------------------------------------------------
>
> Key: SOLR-15150
> URL: https://issues.apache.org/jira/browse/SOLR-15150
> Project: Solr
> Issue Type: New Feature
> Security Level: Public(Default Security Level. Issues are Public)
> Reporter: Chris M. Hostetter
> Assignee: Chris M. Hostetter
> Priority: Major
> Attachments: SOLR-15150.patch
>
>
> When "In-Place" DocValue updates were added to Solr, the choice was made to
> re-use the existing "Atomic Update" syntax, and use the DocValue updating
> code if possible based on the index & schema, otherwise fall back to the
> existing Atomic Update logic (to re-index the entire document). In essence,
> "In-Place Atomic Updates" are treated as a (possible) optimization to
> "regular" Atomic Updates
> This works fine, but it leaves open the possibility of a "gotcha" situation
> where users may (reasonably) assume that an update can be done "In-Place" but
> some aspect of the schema prevents it, and the performance of the updates
> doesn't meet expectations (notably in the case of things like deeply nested
> documents, where the re-indexing cost is multiplicative based on the total
> size of the document tree)
> I think it would be a good idea to support an optional request param users
> can specify with the semantics that say "If this update is an Atomic Update,
> fail to execute it unless it can be done In-Place"
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]