Hi Solr team,
I've discovered a reproducible bug in the HttpJdkSolrClient where the client
becomes permanently blocked after sending concurrent requests with sufficiently
long queries.
Behaviour
When sending 4 or more parallel requests with very long queries the client's
internal task queue fills up completely and the client stops communicating with
Solr entirely.
Conditions to reproduce
*
4+ parallel requests sent simultaneously
*
Query length with more than ~900 characters
*
Using HttpJdkSolrClient
Additional observations
* Sending more than 4 parallel requests within the safe character limit
does not permanently block the client - some requests throw exceptions, but the
client recovers.
* Sending 3 parallel requests with queries well above the threshold also
does not permanently block the client.
* The Solr server version does not appear to affect this behaviour.
Additional observations
* Sending more than 4 parallel requests within the safe character limit
does not permanently block the client - some requests throw exceptions, but the
client recovers.
* Sending 3 parallel requests with queries well above the threshold also
does not permanently block the client.
* The Solr server version does not appear to affect this behaviour.
Workaround
Providing a custom ExecutorService resolves the issue. The specific executor
seems almost irrelevant.
new HttpJdkSolrClient.Builder(DEFAULT_SOLR_URL)
.withConnectionTimeout(CONNECTION_TIMEOUT_MS, TimeUnit.MILLISECONDS)
.withRequestTimeout(REQUEST_TIMEOUT_MS, TimeUnit.MILLISECONDS)
.withExecutor(Executors.newVirtualThreadPerTaskExecutor())
.useHttp1_1(true)
.build()
Demo repository
https://github.com/DarioViva42/solr-blocking-demo/tree/main
Happy to provide additional information or testing.
Best regards,
Dario Viva
