Feature-wise, it's about 3 years behind the features in ES. Although you have to pay for the higher tiers in ES, there's a much richer feature set. The Elastic cloud offering is much nicer than the AWS hosted version, for a similar price.
The opensearch project was released a while ago as opendistro and was just renamed to opensearch a year ago [1]. As for features, it depends what you're doing, but ES upgraded to Lucene 9 6 months ago [2], and they've had document level alerting for at least two years.
I hate the fact that OpenSearch divide in a bipartite the human resources allocated at improving elasticsearch, which is a tragedy.
The hypocrisy is strong at amazon, they have allocated a lot of engineers on OpenSearch and yet their number of contributions pre-fork drama was much much lower (although they were still an important contributor though). But once the management has control and developers can have their own NIH dopamine release, humans are eager to work..
For this human resources reasons, I hope the OpenSearch distraction dies in darkness.
However, beside my utilitaristic goals, I am also one of the only person caring about the technology. And it happens that the technology is generally not found in ElasticSearch but in the much more salient project: Lucene, which it uses.
However, despite its big humans resources, ElasticSearch does a great job at NOT exposing Lucene features and optimizations...
Lucene especially has developped 2 major optimizations the last decade unexposed in ES, one being about shards (don't recall the name), the other being concurrent search. And it happens that the NIH syndrome at amazon allow them to at least reconsider those frozen choices in ES.
https://github.com/opensearch-project/OpenSearch/pull/1500
And I am eager to see the performance implications it will have although the ES approach has some pros too.
In this issue, I descibe what ElasticSearch would do for parallelism, in an ideal world:
https://github.com/elastic/elasticsearch/issues/80693
The division and inevitable inefficiencies are unfortunate, but Elastic.co are the ones who changed ES to use a nasty license. Doesn't this make Elastic.co the jerks?
It sucks it became necessary to fork, but now we have it. Long live FOSS OpenSearch.
Query parallelism (we call it "pseudo sharding") was one of the best things we added to Manticore Search (Elasticsearch alternative). In my opinion Elasticsearch really lacks it. Without that in some use case scenarios (e.g. log analytics which is kind of typical for Elasticsearch) user might require to increase number of shards up to number of CPU cores to get few times lower latency - https://db-benchmarks.com/test-logs10m/#elasticsearch-with-n...
