Description

Allow scripted bulk updates on indices with semantic text fields to determine whether a noop or full reindex of the doc is necessary.

Indexing semantic_text fields is a resource heavy tasl, a common access pattern would therefor be to only update documents if they have not changed.

Imagine a doc under GET /my-index/_doc/1

{ "hash": "SOME-HASH", "semantic": "TEXT" }

The following successfully results in a noop.

POST /my-index/_update/1 { "scripted_upsert": true, "script": { "source": """ if (ctx.op != 'create') { if (ctx._source.hash == params.hash ) { ctx.op = "noop" } } ctx._source = params.doc """, "params": { "hash": "SOME-HASH", "doc": { "hash": "SOME-HASH", "semantic": "TEXT" } } } }

Doing the same through bulk:

POST /semantic-docs-dev/_bulk {"update":{"_id":"1"}} { "scripted_upsert": true, "script": { "source": "if (ctx.op != 'create') { if (ctx._source.hash == params.hash ) { ctx.op = 'noop' } else { ctx._source = params.doc } }", "params": { "hash": "SOME-HASH", "doc": { "hash": "SOME-HASH-2", "semantic": "DIFFERENT TEXT" } } } }

Will result in:

{ "errors": true, "took": 0, "items": [ { "update": { "_index": "semantic-docs-dev", "_id": "/docs/reference/integrations/sonicwall_firewall", "status": 400, "error": { "type": "status_exception", "reason": "Cannot apply update with a script on indices that contain [semantic_text] field(s)" } } } ] }

I presume this is because of the optimizations listed here: https://www.elastic.co/docs/reference/elasticsearch/mapping-reference/semantic-text#semantic-text-updates are harder/impossible to track using scripts.

It would be great to have a more controlled upsert through the bulk api to conditionally update semantic fields.

• Requiring the script to set ctx.semantic_update = true or similar
• Exposing the using new options on bulk Update upserts directly foregoing scripts alltogether.