Description
openedon Oct 27, 2023
I found an issue for /internal/bsearch that I think only affects Discover because we retrieve the full JSON documents: Elasticsearch's response size is dependent on the users' data model, so large avg-sized documents could cause Kibana to struggle when /internal/bsearch holds 500 documents in memory (to compress them and return it to the client).
For context: I recently ran some tests simulating the requests made by Discover in an index populated via makelogs.
The average size of the documents is 9kB, and Discover fetches 500 docs via /internal/bsearch. This results in Kibana using 4.4MB only to handle that request. When we multiply this by multiple concurrent users performing the same request, that can quickly add up (50 users can cause a spike in memory usage of 220MB).
Since Discover is fetching the data (mostly) as-is from ES, we may want to stream the response to avoid overusing the server's resources.
I managed to replicate it locally on main:
- Start ES via
yarn es snapshot - Start Kibana with 550MB max heap size:
NODE_OPTIONS="--max-old-space-size=550" yarn start --no-base-path - While still in a clean state (no data in ES yet), confirm that 100's requests don't OOM Kbana (
source request.sh 200) - Load some data in the index:
yarn makelogs --auth="system_indices_superuser:changeme" --host 127.0.0.1:9200 -d 90/1 -c 10000 - Run 1 request to confirm that now it takes a bit longer to respond (
source request.sh 1) - Run 5 requests and see how Kibana OOMs (
source request.sh 5)
The content of request.sh is below:
for i in {1..$1}
do
curl 'http://localhost:5601/internal/bsearch?compress=true' \
-s -o /dev/null \
-H 'Content-Type: application/json' \
-H 'Authorization: Basic c3lzdGVtX2luZGljZXNfc3VwZXJ1c2VyOmNoYW5nZW1l' \
-H 'elastic-api-version: 1' \
-H 'kbn-version: 8.12.0' \
-H 'x-elastic-internal-origin: Kibana' \
--data-raw '{"batch":[{"request":{"params":{"index":"logstash-*","body":{"sort":[{"@timestamp":{"order":"desc","format":"strict_date_optional_time","unmapped_type":"boolean"}},{"_doc":{"order":"desc","unmapped_type":"boolean"}}],"fields":[{"field":"*","include_unmapped":"true"},{"field":"@timestamp","format":"strict_date_optional_time"},{"field":"relatedContent.article:modified_time","format":"strict_date_optional_time"},{"field":"relatedContent.article:published_time","format":"strict_date_optional_time"},{"field":"utc_time","format":"strict_date_optional_time"}],"size":500,"version":true,"script_fields":{},"stored_fields":["*"],"runtime_mappings":{},"_source":false,"query":{"bool":{"must":[],"filter":[{"range":{"@timestamp":{"format":"strict_date_optional_time","gte":"2022-10-02T22:00:00.000Z","lte":"2023-10-31T16:14:53.238Z"}}}],"should":[],"must_not":[]}},"highlight":{"pre_tags":["@kibana-highlighted-field@"],"post_tags":["@/kibana-highlighted-field@"],"fields":{"*":{}},"fragment_size":2147483647}},"track_total_hits":false,"preference":1696349687154}},"options":{"sessionId":"7feb5c38-dbd2-4c20-83d6-a1ac9e54571$i","isRestore":false,"strategy":"ese","isStored":false,"isSearchStored":false,"executionContext":{"type":"application","name":"discover","url":"/jdm/app/discover","page":"app","id":"new","description":"fetch documents"}}},{"request":{"params":{"index":"logstash-*","body":{"aggs":{"0":{"date_histogram":{"field":"@timestamp","calendar_interval":"1w","time_zone":"Europe/Madrid","min_doc_count":1}}},"size":0,"fields":[{"field":"@timestamp","format":"date_time"},{"field":"relatedContent.article:modified_time","format":"date_time"},{"field":"relatedContent.article:published_time","format":"date_time"},{"field":"utc_time","format":"date_time"}],"script_fields":{},"stored_fields":["*"],"runtime_mappings":{},"_source":{"excludes":[]},"query":{"bool":{"must":[],"filter":[{"range":{"@timestamp":{"format":"strict_date_optional_time","gte":"2022-10-02T22:00:00.000Z","lte":"2023-10-31T16:14:53.238Z"}}}],"should":[],"must_not":[]}}},"preference":1696349687154}},"options":{"sessionId":"7feb5c38-dbd2-4c20-83d6-a1ac9e54571$i","isRestore":false,"strategy":"ese","isStored":false,"isSearchStored":false,"executionContext":{"type":"application","name":"discover","url":"/jdm/app/discover","page":"app","id":"new","description":"fetch chart data and total hits","child":{"type":"lens","name":"lnsXY","id":"unifiedHistogramLensComponent","description":"Edit visualization","url":"/jdm/app/lens#/edit_by_value"}}}}]}' &
done
wait