Qdrant ES Gateway
An Elasticsearch-compatible application-search gateway backed by the Qdrant engine.
The project is aimed at teams that want to keep the Elasticsearch client and request shapes already embedded in an application while using Qdrant for the search and storage engine underneath.
Start here
This is an application-search gateway, not a drop-in replacement for Kibana, log analytics, Lucene scoring, or every Elasticsearch administration API.
Posts
-
Qdrant 1.20.0 through the gateway: almost identical read performance and an auth upgrade check
-
Script fields cannot be empty promises
-
Hit concurrency metadata cannot be invented
-
Requested doc values cannot silently disappear
-
Which dry plots need an irrigation inspection?
-
Query profiles cannot be empty
-
Refreshing the production stack without opening old volumes
-
Score explanations must explain the score
-
Highlight snippets cannot vanish
-
Which batteries belong on the dispatch shortlist?
-
AWS Retail Demo Store: compatible search requests, different trade-offs
-
Search controls cannot be decoration
-
An empty alias batch is not success
-
Testing a cache before building it
-
Find the next page across the whole catalogue
-
Two gateways need one view of their mappings
-
Repeated searches get faster; fresh queries still cost more
-
Rustls security fixes landed
-
Reusing catalog connections helps misses and writes, with a warm-read warning
-
Removing request-local SQLite reconstruction: modest gains, one worse tail
-
One Qdrant search thread improves tails, but hurts typical cold queries
-
Missing alias removals are not success
-
Measure the request timeline before rewriting JSON handling
-
Fewer requests did not make this query faster
-
Can I borrow the large-print edition today?
-
Inside a cold search: a backend diagnostic, not another engine benchmark
-
Sorting twenty records should not fetch ten thousand
-
Keep your sparse token IDs intact
-
Keep coordination durable, let response-cache fills be disposable
-
Asking for less should not make sorting slower
-
Changing transport was not enough
-
A reporting day is not always 24 hours
-
One fewer catalog request, without a proven latency win
-
The cheapest item may be on the next page
-
A shared catalog is not a snapshot
-
Index names are a client contract
-
Alias options must not disappear
-
Who is still waiting for a first reply?
-
Creating an index is not a mapping update
-
Alias add is not a retarget
-
HEAD must not hide storage failures
-
A missing point is not a failed upstream
-
Two passengers, one cancelled departure
-
Source filters should not leak sibling fields
-
Integer hit totals need the minus-one sentinel
-
Disabled hit totals should skip the count
-
Search-after needs more than one page
-
OpenSearch wildcards keep the leading star
-
Is this coastal clip cleared for the UK edit?
-
Invalid pages should not query Qdrant
-
Inner hits need the requested name
-
Unsupported post filters must fail closed
-
A post-filter is not an aggregation filter
-
Packed, but where is the tracking number?
-
Keyword multi-fields need one route
-
A page is not a total
-
Sort syntax must not change the order
-
An evening class with room for one more
-
Arrays are multi-valued fields, not scalar mismatches
-
Optional should really is optional
-
Nested negation keeps its boundary
-
Pattern counts must scan the source
-
The pump is fixed. Take it off the queue.
-
Exists means not empty
-
Zero results still need a positive scroll
-
Mapping updates must not erase the map
-
Invalid UTF-8 is not source data
-
Alias batches are one routing change
-
Bulk mutations need document IDs
-
Bulk action metadata is a write boundary
-
Unchanged updates should not write
-
Partial updates should not erase sibling fields
-
Bulk items should preserve the operation result
-
A partial document is not an upsert source
-
A partial update must preserve the document
-
Accurate bulk index results without per-item reads
-
Standalone create now preserves existing documents
-
Index results should describe the write
-
One index, one collection namespace
-
Bulk create must not become upsert
-
An alias must route the request
-
0.1.2: safer defaults for production traffic
-
A missing update is not an upsert
-
A missing delete is not a successful delete
-
A timeout is a feature
-
Oversized requests are not bad JSON
-
A missing document is a 404
-
One replica is the safe default
-
No phantom indices after creation failures
-
A slash is part of the document ID
-
When bulk errors did not mean errors
-
Three nodes, one disappears, and the search keeps moving
-
Making the bulk request limit real
-
Keep the Elasticsearch client, change the engine
-
Two collections, one stateless gateway
-
Pattern queries without a client rewrite
-
Batching searches and paging through ordered results
-
Compatibility includes the boring lifecycle calls
-
Launch video: keep the Elasticsearch client, change the engine
-
A dress, a tea bag, and a null
subscribe via RSS