[
https://issues.apache.org/jira/browse/SOLR-11677?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16267275#comment-16267275
]
Hoss Man commented on SOLR-11677:
---------------------------------
Just to be clear: in-place updates were always designed (and documented) from
the stand point of being a _potential_ performance optimization that may happen
when possible on a subset of atomic-updates (that supports it). If the schema
doesn't support Atomic Updates, or a situation occurs where a "regular" Atomic
Update may lose data (because of re-indexing), then users (today) should
certainly not attempt an in-place update -- that was the entire point of not
having special syntax! from a user perspective they should treat it just like
any other atomic update -- but in the right circumstance, some atomic-updates
will be done "in place"
bq. So there should really be a way to enforce in-place docvalues updates and
fail otherwise!
I guess I don't see any downside to this? .. perhaps something along the lines
of...
{code}
{
"id":"mydoc",
"price":{"set":99,
"inplace":true},
"popularity":{"inc":20}
}
{code}
...where the value of "inplace" can be {{true}}, {{false}}, or {{null}} ...
* if {{true}}: the update will fail unless it can be done in in place
* if {{false}}: the update will force a reindex regardless of wether or not it
would be possible to be done in place
* if {{null}} *(default)*: the update will be done in place if possible,
otherwise reindexing will be done
the most common case being that nothing is specified and {{null}} is used by
default; the most likeley un-common case being a user specifies
{{inplace:true}} (while {{inplace:false}} would probably only ever exist in
tests, but might prove useful down the road)
> Add ability to "enforce" in-place updates
> -----------------------------------------
>
> Key: SOLR-11677
> URL: https://issues.apache.org/jira/browse/SOLR-11677
> Project: Solr
> Issue Type: Improvement
> Security Level: Public(Default Security Level. Issues are Public)
> Components: update
> Affects Versions: 6.6
> Reporter: Uwe Schindler
>
> There is currently a problem with in-place updates: Under certain conditions,
> an in-place update does not work, so Solr internally reindexes the whole
> document. This is not bad in most cases, unless you have some fields in your
> documents that are not stored (e.g. large fulltext fields). In that case, the
> try to do in-place update leads to data loss!
> I'd suggest to add a request parameter in the update to "enforce in-place
> mode". If not all conditions are met (e.g. field configuration is
> incompatible, version field does not have all required settings, or user uses
> other operations than "add" or "increment"), the request should simply fail -
> to prevent data loss!
> There is also a related issue that may lead to data loss because of the same
> issue: SOLR-11286. In this case it is completely unexpected because
> everything is correct from the schema settings, but it still does reindex.
> So there should really be a way to enforce in-place docvalues updates and
> fail otherwise!
> Another thing to maybe think of: Only prevent "reindexing" updates if
> documents have some fields that are unstored!
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
