App Search on Elastic Cloud (out of the box) seems slow: 800ms ~ 1200ms

Hi,

We have an AppSearch deployment running on Elastic Cloud Azure Netherlands. From here we use the crawler to crawl "https://www.unive.nl". The crawler is configured to crawl by use of the sitemap with a crawl depth of 1. This results in 2155 indexed documents.

For search we hit the search API with a query including "search_fields" and "result_fields" settings.
Response times are on average between 800ms and 1,2s. This seems a but slow to us. And our customer is complaining.

We tried increasing AppSearch deployment resources (from 4GB to 8GB per instance), but this didn't help. Search latency in ES is 1ms which seems fine. We tried search from different points in network, but it doesn't make a difference. Quering Elasticsearch directly for the same data takes around 200 ms. Does AppSearch introduce extra latency of around 600 ~ 1000 ms?

Do you agree it's slow? What else can we do to speed it up?

Some screenshots with more detail on the setup attached.

Thank you!

Kind regards,
Mark Niemeijer
P.S. The query we use:

{
    "query": "tand",
    "page": {
        "size": 100
    },
    "search_fields": {
        "title": {},
        "body_content": {},
        "meta_keywords": {},
        "meta_description": {},
        "url": {}
    },
    "result_fields": {
        "url": {
            "raw": {
                "size": 500
            }
        },
        "title": {
            "raw": {
                "size": 100
            }
        },
        "body_content": {
            "snippet": {
                "size": 100,
                "fallback": true
            }
        }
    }
}

P.P.S. some screenshots:


image
image
image

1s response. TTFB +900ms
image

ES latency:
image

1 Like

Hi,

any suggestions? I could really use some help here. Customer is thinking of moving away from Elastic Cloud to dedicated hardware in own data center.

Thanks.

Kind regards,
Mark

Sorry you're having issues with your App Search deployment!

To help us better isolate the source of the slowness, can you make an API call to your deployment and share the value of the X-Request-Id header from it? That should give us the time it takes to execute the actual query on the server, without SSL handshake and other networking-related overhead from your machine to the cluster itself.

Hi,

thanks for responding! Please find screenshot below:

Hi @oleksiy-elastic by any chance did you find the time to look at the screenshot? Thanks!

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.