Evaluating Vector Databases as AppCat services #177
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,132 @@ | ||
| = ADR 0051 - Evaluating vector databases as AppCat services | ||
| :adr_author: Simon Beck | ||
| :adr_owner: Schedar | ||
| :adr_reviewers: | ||
| :adr_date: 2026-01-14 | ||
| :adr_upd_date: 2026-01-14 | ||
| :adr_status: draft | ||
| :adr_tags: database,service | ||
|
|
||
| include::partial$adr-meta.adoc[] | ||
|
|
||
| [NOTE] | ||
| .Summary | ||
| ==== | ||
| Exisitng AppCat services already provide some vector capabilities. | ||
| For any other use-cases the required databases should be provided as solutions. | ||
| ==== | ||
|
|
||
| == Context | ||
|
|
||
| More customers require vector databases for various AI related workloads. | ||
|
|
||
| To provide the best services possible some of the most popular open source solutions are evaluated in this ADR. | ||
|
|
||
|
|
||
| Solutions being evaluated: | ||
|
Member
There was a problem hiding this comment. How do these compare performance and resource consumption wise?
Contributor
Author
There was a problem hiding this comment. I did not do any benchmarking. Would also have been a pain since every solution has their own API... Chroma and qdrant were basically negligible, with less than 100mb memory and 1mCPU. Milvus with all its bundled services was by far the heaviest on resources even in idle. I did not take any note of the concrete numbers for that though. |
||
|
|
||
| - OpenSearch/ElasticSearch | ||
| - PostgreSQL with pgvector | ||
| - Redis | ||
| - Chroma | ||
| - Qdrant | ||
| - Milvus | ||
|
|
||
| === OpenSearch/ElasticSearch | ||
|
|
||
| OpenSearch/ElasticSearch provides vector capabilities https://opensearch.org/platform/vector-search/[out of the box]. | ||
| Either one of these products are already on the roadmap for AppCat. | ||
|
|
||
| === PostgreSQL with pgvector | ||
|
|
||
| StackGres and CNPG both already provide the pgvector extension. | ||
| For StackGres it needs to be installed via the `extensions` stanza. | ||
| CNPG pre-installs it in their default image. | ||
| So the vector capabilities of our PostgreSQL offering is already there. | ||
| Additional documentation to make users aware of this feature is recommended though. | ||
|
|
||
| === Redis | ||
|
|
||
| Redis also provides vector capabilities out of the box. | ||
| Redis provides documentation on how to use it: https://redis.io/docs/latest/develop/get-started/vector-database/#prepare-the-demo-dataset | ||
|
|
||
| === Dedicated Vector databases | ||
|
|
||
| [cols=4] | ||
| |=== | ||
| |Criteria | ||
| |Chroma | ||
| |Qdrant | ||
| |Milvus | ||
|
|
||
| |Installation | ||
| |https://github.com/amikos-tech/chromadb-chart[community helm chart], but very limited functionality, missing HA | ||
| |Official Helm chart available, enterprise operator | ||
| |Official chart and operator available | ||
|
|
||
| |Backup | ||
| |https://cookbook.chromadb.dev/strategies/backup/[Export tool available, filesystem snapshots and filesystem snapshots] | ||
| |https://qdrant.tech/documentation/database-tutorials/create-snapshot/[Pretty involved process for clusters] | ||
| |https://milvus.io/docs/milvus_backup_cli.md[Backup tooling available] | ||
|
|
||
| |Metrics | ||
| |https://docs.trychroma.com/guides/deploy/observability[Opentelemetry traces] | ||
| |https://qdrant.tech/documentation/guides/monitoring/[OpenMetrics/Prometheus] | ||
| |https://milvus.io/docs/monitor.md[OpenMetrics/Prometheus] | ||
|
|
||
| |Maintenance/2nd-day operations | ||
| |https://cookbook.chromadb.dev/running/maintenance/[Chroma ops tool to do maintenance on the database] | ||
| |https://qdrant.tech/documentation/guides/administration/[Administration guide] | ||
| |Administration guides available that cover a lot of topics | ||
|
|
||
| |Update/Upgrade | ||
| |https://docs.trychroma.com/docs/overview/migration[Migration guide available], recent rewrite in rust, rewrite doesn't have authentication | ||
| |https://qdrant.tech/documentation/faq/qdrant-fundamentals/?q=upgradeupgrade+guide#how-do-i-avoid-issues-when-updating-to-the-latest-version[Updates to the next consecutive version are supported] | ||
| |Automated via their operator | ||
|
|
||
| |License | ||
| |Apache License 2.0 | ||
| |Apache License 2.0 | ||
| |Apache License 2.0 | ||
|
|
||
| |=== | ||
|
|
||
| === Chroma | ||
| Chroma's maturity doesn't seem to be there, many things are not documented on the official page but rather on a community driven page. | ||
| Only a rudimentary helm chart is available. | ||
| The recent rewrite in rust dropped features that were previously available. | ||
|
|
||
| === Qdrant | ||
| Qdrant's documentation is excellent. | ||
| However, there's a lot of "available on our cloud offering" there as well for features like cluster backups, which might be concerning for future anti-consumer decisions. | ||
| The overall architecture and concepts are very inspired by ElasticSearch. | ||
|
|
||
| === Milvus | ||
|
|
||
| Milvus is by far the most complex solution of the three. | ||
| Although it provides an operator to handle the deployment, the amount of services it needs and deploys is extensive: | ||
|
|
||
| - dedicated etcd | ||
| - message queue | ||
| - minio | ||
| - zookeeper | ||
|
|
||
| This rivals Kafka in complexity. | ||
|
|
||
| == Decision | ||
|
|
||
| The AppCat portofolio already contains two existing services that provide vector capabilities. | ||
| OpenSearch/ElasticSearch is on the roadmap. | ||
| These services and products are very well understood and widely used. | ||
| AppCat should provide documentation to make users aware of the vector capabilities of these services. | ||
|
|
||
| The dedicated vector databases might fit some use cases better, but of the three concrete requests we got, three different products were mentioned or requested. | ||
| Given the initial effort to make a service AppCat managed, it's not feasible to add these services just for 1-2 instances. | ||
| In these cases the databases should be handled as a solution. | ||
| If at one point the adoption and demand for a specific product increases, the need to provide it as a self-service AppCat service can be re-evaluated. | ||
|
|
||
| == Consequences | ||
| We do not adopt any new vector database services into the AppCat service catalog. | ||
| Instead we make users aware about the vector capabilities of the existing services. | ||
|
Member
There was a problem hiding this comment. How would customers make use of these capabilities? I'm assuming there will still be some initial implementation needed by us to allow the use of it. |
||
|
|
||
| For any specific vector database deploying it as a solution should be considered. | ||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1,5 +1,5 @@ | ||
| { | ||
| "adr_number": "0049", | ||
| "full_name": "VSHNeer Name", | ||
| "adr_title": "Title", | ||
| "adr_reviewers": "", | ||
|
|
||
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Are there any specific features required or did customers only inquire about vector databases in general?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The later.