mrkeyoor.comAsk

seaweedfs/seaweedfstested 03 Oct 2026commit 3e679e9

seaweedfs review

Our run of SeaweedFS installed 1,160 packages and built successfully, but only 14 of 31 test results passed before the 900-second cap, so it belongs with operators who can reproduce its full integration environment.

  • Install 224 s
  • Build 313 s
14/31passed, run exited non-zero
  • 1,160 packages
Self-Hostedevaluationupdated 04 Oct 2026
+86stars / 7d

SeaweedFS is a distributed storage system that serves S3 objects, ordinary files, and Iceberg or Lance tables from the same data. It is built for operators who have too many small files for a conventional directory tree and want to add capacity by starting more volume servers.

Verdict

Our run of SeaweedFS installed 1,160 packages and built successfully, but only 14 of 31 test results passed before the 900-second cap, so it belongs with operators who can reproduce its full integration environment. Use it when small-file storage must serve several protocols and you are ready to own balancing and recovery. Choose a narrower object store when S3 alone solves the problem.

We ran it

Lab card: what happened when we ran seaweedfs
Lab card from our sandbox run
Screenshot of seaweedfs (seaweedfs.com)
seaweedfs.com
Install✓ · 224s1160 packages
Build✓ · 313s
Tests✗ timed out · 900s14 passed · 17 failed of 31 (go test)
Repo4125 files~1,001,038 lines of source · 177.3 MB · 74 CI workflows · tests dir

Answers from our run

Does seaweedfs build from source?

Dependencies installed in 224 seconds (1160 packages), and the build succeeded in 313 seconds. We cloned commit 3e679e9 into a clean Debian container with 3 CPUs and no project-specific setup.

Do seaweedfs's tests pass?

No. The project's own test command finished in 900 seconds and reported 14 passed, 17 failed (go test).

Who should not use seaweedfs?

Teams that want storage they can deploy and ignore: the README says balancing, vacuum, erasure coding, and repair run on demand.

What are the alternatives to seaweedfs?

Ceph, RustFS, Garage. Our run of SeaweedFS installed 1,160 packages and built successfully, but only 14 of 31 test results passed before the 900-second cap, so it belongs with operators who can reproduce its full integration environment.

Setup3/5One binary starts fast; our 900-second suite did not finish cleanly
Docs5/5Clear starts for one node, Docker, Helm, source, and scale-out
Community5/535,234 stars, an October 4 push, and same-day issue activity
Maturity4/5Release 4.48 is active; our full test run still failed

Discussed on

  1. hnSeaweedFS fast distributed storage system for blobs, objects, files and datalake328 points
  2. hnSeaweedFS – A simple and highly scalable distributed file system119 points
  3. hnSeaweedFS – A simple and highly scalable distributed file system with S3 API106 points
  4. hnSeaweed-FS: simple distributed file system written in Go67 points
  5. hnSeaweedfs10 points

Who it’s for

Infrastructure teams storing very large numbers of small files or objects.
Operators who need S3, FUSE, WebDAV, or SFTP access to the same storage.
Data teams that want S3 Tables and an Iceberg REST catalog beside object storage.
Self-hosters prepared to own replication, metadata, balancing, repair, and recovery.

Who it’s NOT for

Teams that want storage they can deploy and ignore: the README says balancing, vacuum, erasure coding, and repair run on demand.
Buyers who require a clean full test result in a fresh container: our 900-second run reported 14 passed and 17 failed out of 31.
Small applications that only need a basic S3 endpoint: SeaweedFS also brings master, volume, filer, and gateway roles.
Operators unwilling to test an integration environment: the failing log ended with S3 tagging requests unable to list or create buckets through localhost.

Setup reality

Our fresh Debian sandbox installed commit 3e679e9 in 224 seconds, adding 1,160 packages. The build passed in 313 seconds. Tests timed out at 900 seconds: Go reported 14 passed and 17 failed out of 31, with the final failures in S3 tagging tests.

The one-process start needs a data directory plus S3 access and secret keys. A production deployment adds master and volume topology, replication settings, filer metadata, persistent disks, gateway credentials, monitoring, and backups. The README's Kubernetes example uses 3 masters, 3 volume replicas, 2 filer replicas, and an authenticated S3 service.

Capacity can grow by adding a volume server, but the README says data does not rebalance until an operator asks. Our checkout had 74 CI workflow files and a tests directory, yet no Dockerfile, even though the project publishes a container image.

One binary serves objects, files, and tables

The 177.3 MB checkout puts a master, volume server, filer, and several gateways behind one weed binary. The master tracks volumes instead of every file, and clients can contact the volume server holding a blob without sending each read through the master. Small files are packed into append-only volume files. Gateways expose the same data through S3, WebDAV, SFTP, FUSE, and other interfaces.

That breadth is also visible in the 4,125 files and about 1,001,038 lines of source. One weed mini process can start a master, volume server, filer, S3 endpoint, WebDAV service, Iceberg REST catalog, and admin interface. It is a convincing way to try the system because you can copy an object into a bucket without assembling a cluster first.

The single process should not set expectations for production. SeaweedFS splits responsibility among masters, volume servers, filers, metadata stores, and gateways. Its Helm example uses 3 master replicas, 3 volume replicas, and 2 filer replicas before counting the S3 service. Teams must choose replication placement, persistent storage, credentials, TLS, monitoring, backup, and a repair routine.

New volume servers do not rebalance existing data

The README's production example assigns 500 GiB to each of 3 volume replicas, and adding another server still does not rebalance existing data automatically. Balancing, vacuum, erasure coding, and repair can run from weed shell or the maintenance worker. That keeps policy in the operator's hands and puts the result there too.

The data path is attractive for small-file workloads. A volume server keeps a 16-byte index entry per blob in memory, and the project says file content takes one disk read, including erasure-coded data. Those are architecture claims from the project, not measurements from our sandbox. We did not test throughput, latency, disk failure, or file-count limits, so the decision should rest on a pilot using your disks and object sizes.

The filer widens the design space again. It can use SQLite for a small setup or external stores such as PostgreSQL, MySQL, Redis, and several distributed databases. That choice affects the namespace's availability, backup, and restore path even when the volume data is healthy. An S3-only team may find this freedom expensive. A team already operating one of those stores may prefer it to a fixed metadata layer.

What happened when we ran it

Our measurement setup was a fresh unprivileged Debian container with 3 CPUs, 8 GB of RAM, Go 1.24, and no secrets. We cloned commit 3e679e9. We measured 224 seconds for installation, which added 1,160 packages, and built the project in another 313 seconds. The checkout included 74 CI workflow files and a tests directory, while our scan found no Dockerfile.

We ran the full test command until the 900-second limit. Go reported 14 passed and 17 failed out of 31. The last failures were under test/s3/tagging. Cleanup could not list buckets, and a versioned-object test could not create a bucket after 3 attempts. The error showed an S3 request to a localhost URL with status code 0. The supplied log does not show why that local endpoint could not answer.

This result does not prove that object tagging is broken. It does prove that the repository's full test command did not complete cleanly in our fresh container. Reproducing the expected integration services is part of evaluating SeaweedFS from source. A storage team should get the S3 tests for its intended features passing, then add node-loss and restore drills before trusting irreplaceable data.

S3 shares the system with file and table interfaces

Release 4.48 supports familiar S3 behavior including versioning, Object Lock, lifecycle rules, tagging, multipart uploads, policies, IAM, and STS. The README lists 73 bucket and object operations, 36 S3 Tables operations, 39 IAM operations, and 5 STS operations. That is wide coverage, though a checklist cannot show whether one application's exact request patterns behave like AWS.

FUSE and WebDAV clients can reach the same filer namespace, while Iceberg and Lance catalogs sit beside the object API. This combination is the project's best reason to exist. If one data set must serve application uploads, mounted files, and analytical tables, fewer storage copies can simplify the surrounding system. If you need only object storage, every extra interface is another behavior to secure, monitor, and test.

Release 4.48 and October activity show active maintenance

GitHub showed 35,234 stars and 770 open issues and pull requests combined on October 4, 2026. The repository was pushed that day, and current issues and pull requests were updated that day as well. Release 4.48 arrived on September 28 after four other releases since August 22. The project is moving quickly, so pin releases and read upgrade notes before putting a cluster under automated deployment.

SeaweedFS is a strong candidate when small files, S3 clients, mounted access, and table catalogs genuinely need one storage layer. Our 313-second successful build shows the source is buildable in a plain container, while 17 failed results show that validation needs more than the quick start. The last useful test is operational: remove a node, interrupt metadata, restore from backup, and see whether your team can explain every outcome.

Alternatives

ProjectWhat it isPick it when
CephA distributed platform for object, block, and file storage.pick this instead when block storage and an established multi-protocol storage stack are required.
RustFS gh↗An S3-compatible object store written in Rust.pick this instead when S3 is the job and SeaweedFS's filer and table features add unwanted scope.
GarageAn S3-compatible store aimed at small self-hosted, geographically distributed deployments.pick this instead when a smaller geo-distributed S3 system fits better than SeaweedFS's broader file stack.

What people are saying

  1. [github-trending] seaweedfs/seaweedfs

Sources

  1. SeaweedFS README
  2. SeaweedFS repository
  3. SeaweedFS release 4.48
  4. SeaweedFS open issues and pull requests

More self-hosted reviews