Thanks, Robert This seems much more concrete than just changing the licensing since it seems like we actually get new functionality from the test layer with the switch.
I'm generally hesitant about swapping test infra just to use a differently licensed project which was my main concern about just switching MinIO for RustFS. This was especially worrying to me since RustFS needed upstream fixes and workarounds before it reached parity. Floci feels different to me because the argument is about coverage, not licensing. Today we run credential-vending tests against MinIO with kmsUnavailable=true because MinIO's STS policy evaluator doesn't handle KMS ARNs the way AWS does. If a Floci spike can exercise STS AssumeRole, session-policy enforcement, and KMS-related policy paths without that workaround, that's a real improvement in what we're validating. I'm supportive of steps 1–3 as a low-risk evaluation: add a Floci testcontainer next to the existing MinIO helper, port one or two credential-vending tests, and explicitly check the behaviors you listed. If the existing MinIO tests still pass unchanged, and Floci adds coverage we currently skip or approximate that's a success for me. Net: I'm not in favor of ripping out MinIO today, but I am in favor of the spike you outlined. If it closes the KMS/policy gap and doesn't add flake, I'd be much more open to migrating the STS/IAM-heavy tests than I would be to another license-motivated swap. On Tue, Jun 30, 2026 at 5:35 AM Robert Stupp <[email protected]> wrote: > I would separate LocalStack from MiniStack here. > > For LocalStack, my concern is mostly operational/project risk rather than > technical capability. > Recent LocalStack releases require auth tokens/account setup, including for > CI, and event reporting also needs to be explicitly disabled. > That may be fine for many companies, but it is awkward as a default > dependency for ASF CI and for contributors running the test suite locally. > It reintroduces account, token, or entitlement handling into tests > that should ideally be clone-and-run. > > MiniStack looks like a much better candidate on that axis: it's MIT > licensed, requires no account/API key, has no telemetry, and is > AWS-focused. > I would not object to trying it. > > The reason I suggested Floci first is that it seems to cover the exact > shape of the Polaris tests. > It also has a Testcontainers module and advertises very fast startup, which > matters for CI. > The broader Floci family may also become useful later because Polaris has > GCP and Azure storage paths too; MiniStack is AWS-only. > > So my preference would be: > > 1. Do a small Floci spike for one or two current MinIO STS/IAM tests. > 2. Define pass/fail criteria around the behavior we actually need: > AssumeRole, > vended credentials, denied operations, session policies, path-style and > virtual-host-style behavior. > 3. If Floci misses important behavior, try the same spike with MiniStack. > 4. Avoid LocalStack as the default unless we decide the auth/token/CI setup > is > acceptable for the project. > > That keeps this as an evidence-based test replacement discussion, not a > broad tool endorsement. > > On Tue, Jun 30, 2026 at 11:33 AM Alexandre Dutra <[email protected]> > wrote: > > > Hi Robert, > > > > Floci looks very promising. I agree we should try it out. Compared to > > LocalStack, it's 100% free, and compared to Ministack, it also covers > > Azure and GCS, so it seems really a good fit for our testing use > > cases. > > > > Thanks, > > Alex > > > > On Tue, Jun 30, 2026 at 8:25 AM Jean-Baptiste Onofré <[email protected]> > > wrote: > > > > > > Hi Robert, > > > > > > Thanks for sharing this. I wasn't familiar with Floci, but it looks > very > > > interesting. > > > > > > Since it is MIT licensed, it’s a good fit from a legal standpoint. It > > also > > > seems like a great candidate for our integration tests, especially > given > > > that it functions more as an AWS emulator than just an object store. > > > > > > Regards, > > > JB > > > > > > > > > On Mon, Jun 29, 2026 at 7:47 PM Robert Stupp <[email protected]> wrote: > > > > > > > Reviving this thread with one more option, specifically for the test > > side. > > > > > > > > I agree with the distinction made earlier in the thread: the best > > backend > > > > for getting-started examples and the best backend for integration > > tests do > > > > not have to be the same thing. > > > > > > > > For tests that currently rely on MinIO because they need S3 plus > > STS/IAM, I > > > > think Floci [1] [2] is worth evaluating as a replacement: > > > > > > > > Unlike RustFS, Garage, or other S3-compatible stores, Floci is an AWS > > > > emulator rather than only an object-store implementation. > > > > That means the interesting surface for these tests is not just S3 > > CRUD, but > > > > S3 together with IAM, STS, session credentials, policy evaluation, > > KMS, and > > > > Secrets Manager. > > > > That seems closer to what Polaris needs for credential vending and > > related > > > > cloud integration coverage. > > > > > > > > I would not suggest switching everything in one step. A safer > migration > > > > plan could be: > > > > > > > > 1. Add a small Floci-backed testcontainer/helper next to the existing > > MinIO > > > > helper. > > > > 2. Port one or two MinIO-based credential-vending tests to Floci > > without > > > > removing the MinIO tests. > > > > 3. Explicitly check the behavior we care about: > > > > - STS AssumeRole works with vended credentials > > > > - session policies are enforced > > > > - denied S3 operations fail where expected > > > > - path-style and virtual-host-style access behavior is understood > > > > - existing MinIO-specific workarounds, such as KMS limitations, > are > > > > either > > > > still needed or can be removed for Floci > > > > 4. If the coverage is at least equivalent, migrate the remaining > > STS/IAM > > > > MinIO tests. > > > > 5. Only then remove the MinIO testcontainer usage that is no longer > > needed. > > > > > > > > There may also be a broader testing upside if the AWS spike works. > > > > The Floci suite is not limited to AWS: > > > > - floci-gcp [3] claims Cloud Storage, IAM, Secret Manager, and Cloud > > KMS > > > > support, and > > > > - floci-az [4] claims Blob Storage and Key Vault support. > > > > > > > > All three advertise very fast startup / low idle memory, so this > could > > > > reduce test duration versus heavier service containers. > > > > > > > > I would keep that as a follow-up benefit rather than the first > > migration > > > > goal, though. > > > > For any later Azure/GCP migration, we should make provider-specific > > auth > > > > part of the spike rather than assuming parity: > > > > > > > > - AWS: STS AssumeRole, session policies, IAM allow/deny behavior, > > > > KMS/Secrets Manager where relevant. > > > > - GCP: Cloud Storage plus IAM, Secret Manager, and KMS behavior used > by > > > > Polaris. > > > > - Azure: Blob Storage plus Key Vault, and explicit SAS-token coverage > > > > because > > > > SAS support is not prominent in the top-level floci-az docs. > > > > > > > > This would keep the quickstart discussion separate. > > > > Ozone may still be the better long-term fit for examples if/when its > > STS > > > > story is complete enough, especially because it is an ASF project. > > > > But for CI/integration tests that need an AWS-shaped local service, > > Floci > > > > seems like a strong candidate to evaluate. > > > > > > > > Robert > > > > > > > > [1] https://floci.io/ > > > > [2] https://github.com/floci-io/floci > > > > [3] https://github.com/floci-io/floci-gcp > > > > [4] https://github.com/floci-io/floci-az > > > > > > > > > > > > On Tue, Jan 27, 2026 at 9:05 AM Yong Zheng <[email protected]> > > > > wrote: > > > > > > > > > sounds good. > > > > > > > > > > Thanks, > > > > > Yong > > > > > > > > > > > On Jan 26, 2026, at 9:03 PM, Yufei Gu <[email protected]> > > wrote: > > > > > > > > > > > > Thanks a lot for digging deep on it, Yong! Given that virtual > host > > > > > style is > > > > > > the default for AWS s3[1], maybe we will still need it for s3 > > > > compatible > > > > > > storage in the long term. I think we could work with the RustFS > > > > community > > > > > > to get it resolved, while gradually migrating from MinIO to > RustFS > > for > > > > > > tests. > > > > > > > > > > > > 1. > > > > > > > > > > > > > > > > > > https://github.com/apache/polaris/blob/0b54f7046295ff19d434f9f0bd47b0b612be51a5/spec/polaris-management-service.yml#L1146 > > > > > > > > > > > > Yufei > > > > > > > > > > > > > > > > > >> On Mon, Jan 26, 2026 at 7:05 AM Yong Zheng <[email protected]> > > wrote: > > > > > >> > > > > > >> Hello, > > > > > >> > > > > > >> So in MinioContainer.java, we have following: > > > > > >> ``` > > > > > >> private S3Client createS3Client() { > > > > > >> return S3Client.builder() > > > > > >> .httpClientBuilder(UrlConnectionHttpClient.builder()) > > > > > >> .applyMutation(builder -> > > > > > >> builder.endpointOverride(URI.create(s3endpoint()))) > > > > > >> .applyMutation(builder -> region.ifPresent(r -> > > > > > >> builder.region(Region.of(r)))) > > > > > >> // > .serviceConfiguration(s3Configuration(s3PathStyleAccess, > > > > > >> s3UseArnRegionEnabled)) > > > > > >> // credentialsProvider(s3AccessKeyId, s3SecretAccessKey, > > > > > >> s3SessionToken) > > > > > >> .credentialsProvider( > > > > > >> > > > > > >> > > > > > > StaticCredentialsProvider.create(AwsBasicCredentials.create(accessKey(), > > > > > >> secretKey()))) > > > > > >> .build(); > > > > > >> } > > > > > >> ``` > > > > > >> > > > > > >> So we are not enforcing one or the other for access style which > it > > > > will > > > > > be > > > > > >> default (which if virtual host style access based on what i > > recalled). > > > > > >> > > > > > >> Thus, with simple changes I mentioned above, the test cases with > > path > > > > > >> style will work (PolarisRestCatalogMinIOIT.java) as we are > setting > > > > > >> setPathStyleAccess to true implicitly: > > > > > >> ``` > > > > > >> protected StorageConfigInfo getStorageConfigInfo() { > > > > > >> AwsStorageConfigInfo.Builder storageConfig = > > > > > >> AwsStorageConfigInfo.builder() > > > > > >> .setStorageType(StorageConfigInfo.StorageTypeEnum.S3) > > > > > >> .setPathStyleAccess(true) > > > > > >> .setEndpoint(endpoint) > > > > > >> > .setAllowedLocations(List.of(storageBase.toString())); > > > > > >> > > > > > >> return storageConfig.build(); > > > > > >> } > > > > > >> ``` > > > > > >> > > > > > >> The problem is RestCatalogMinIOSpecialIT.java where we have > couple > > > > place > > > > > >> where we are testing for virtual host style access such as > > following: > > > > > >> ``` > > > > > >> @ParameterizedTest > > > > > >> @CsvSource("true, true,") > > > > > >> @CsvSource("false, true,") > > > > > >> @CsvSource("true, false,") > > > > > >> @CsvSource("false, false,") > > > > > >> @CsvSource("true, true, VENDED_CREDENTIALS") > > > > > >> @CsvSource("false, true, VENDED_CREDENTIALS") > > > > > >> public void testAppendFiles( > > > > > >> boolean pathStyle, boolean stsEnabled, AccessDelegationMode > > > > > >> delegationMode) > > > > > >> throws IOException { > > > > > >> ``` > > > > > >> > > > > > >> Thus, if it is okay to not have tests cover virtual host, I can > > remove > > > > > >> them and switch to rustfs testcontainers (similar to the minio > > one but > > > > > >> couple lines changes) and do the final test. From getting-start > > > > example, > > > > > >> STS is all working as I am setting vended credentials implicitly > > in > > > > the > > > > > >> spark command ( > > > > > >> > > https://github.com/apache/polaris/tree/main/getting-started/rustfs): > > > > > >> ``` > > > > > >> --conf > > > > > >> > > > > > > > > > > > > spark.sql.catalog.polaris.header.X-Iceberg-Access-Delegation=vended-credentials > > > > > >> > > > > > >> ``` > > > > > >> > > > > > >> Thanks, > > > > > >> Yong Zheng > > > > > >> > > > > > >> > > > > > >>> On 2026/01/26 02:10:10 Dmitri Bourlatchkov wrote: > > > > > >>> Thanks for the analysis, Yong! > > > > > >>> > > > > > >>> I do not think we have any MinIO-based tests that rely on > virtual > > > > > >>> host-stype access... I may be mistaken, though... Did you come > > across > > > > > any > > > > > >>> specific cases like this? > > > > > >>> > > > > > >>> IIRC all MinIO-based examples and tests use path-style access. > > > > > >>> > > > > > >>> Just to confirm, STS (cred. vending) works with RustFS, right? > > > > > >>> > > > > > >>> Thanks, > > > > > >>> Dmitri. > > > > > >>> > > > > > >>>> On Sat, Jan 24, 2026 at 11:12 PM Yong Zheng < > [email protected]> > > > > > wrote: > > > > > >>> > > > > > >>>> So I spent some time on this today. With path style access, > > rustfs > > > > > will > > > > > >>>> work 100% with couple lines changes. However, when dealing > with > > > > > virtual > > > > > >>>> host style access, it is not happy with testcontainer due to > > > > > >> testcontainer > > > > > >>>> map a random port to host and rustfs is very restrict on > server > > > > > domains > > > > > >>>> (could be bug if not intensional). Take an example, when I > have > > > > rustfs > > > > > >>>> mapped to port 9000 and testcontainer maps it to 62226 (random > > > > > selected > > > > > >>>> port), the health check happens on localhost:62226 which is > not > > > > > >> something > > > > > >>>> we can set ahead of time for RUSTFS_SERVER_DOMAINS. Same > > observation > > > > > >> had > > > > > >>>> being reported in https://github.com/rustfs/rustfs/issues/308 > > where > > > > > >>>> others confirm: > > > > > >>>> ``` > > > > > >>>> ...I tried the domain name method again. When configuring > > > > > >>>> RUSTFS_SERVER_DOMAINS, if the domain name port is not 443, you > > need > > > > to > > > > > >> fill > > > > > >>>> in the port as well, for example, test.rustfs.cn:9000. > > > > > >>>> ``` > > > > > >>>> But with testcontainer, this will map to some random port on > > host as > > > > > >>>> describe above then throw InvalidBucketName error. > > > > > >>>> > > > > > >>>> Another issue is https://github.com/rustfs/rustfs/issues/1593 > > where > > > > > >>>> rustfs thinks that mapping mentioned above is a bucket name. > > > > > >>>> > > > > > >>>> That being said, if we want to change our user facing > > > > getting-started > > > > > >>>> example from minio to rustfs, it will work 100%. However, we > > can't > > > > > >> change > > > > > >>>> our integration tests from minio to rustfs if we need to > > continue > > > > > >> cover the > > > > > >>>> tests on virtual host style access. > > > > > >>>> > > > > > >>>> Thanks, > > > > > >>>> Yong Zheng > > > > > >>>> > > > > > >>>> On 2026/01/19 19:36:48 Yufei Gu wrote: > > > > > >>>>> I think it's a good plan to distinguish between > getting-started > > > > > >> examples > > > > > >>>>> and integration tests, esp. if RustFS is much faster. Thanks > > > > Dmitri. > > > > > >>>>> Will current integration tests fail if we switch to RustFS > from > > > > > >> minio? > > > > > >>>>> > > > > > >>>> > > > > > >> > > > > > > > > > > > > https://github.com/apache/polaris/blob/5e5779f115472fa8614a6820b1e178520d927012/integration-tests/build.gradle.kts#L70-L70 > > > > > >>>>> > > > > > >>>>> Yufei > > > > > >>>>> > > > > > >>>>> > > > > > >>>>> On Mon, Jan 19, 2026 at 11:23 AM Dmitri Bourlatchkov < > > > > > >> [email protected]> > > > > > >>>>> wrote: > > > > > >>>>> > > > > > >>>>>> Hi Yufei, > > > > > >>>>>> > > > > > >>>>>> From my POV we need some storage for "getting started" > > examples > > > > and > > > > > >>>> some > > > > > >>>>>> storage for CI. > > > > > >>>>>> > > > > > >>>>>> They may be the same or different. > > > > > >>>>>> > > > > > >>>>>> I think Apache Ozone is preferable for examples (when it > > supports > > > > > >> STS) > > > > > >>>>>> because it's also inside the ASF and examples do not have to > > be as > > > > > >>>> lean as > > > > > >>>>>> containers used in CI. > > > > > >>>>>> > > > > > >>>>>> IMHO, RustFS looks promising for CI due to simplicity / > > > > > >> performance... > > > > > >>>> but > > > > > >>>>>> it remains to be tested :) > > > > > >>>>>> > > > > > >>>>>> As long as the RustFS project remains active and aligned > with > > S3 > > > > > >> API > > > > > >>>>>> changes, I think it does not matter so much that it is not > an > > ASF > > > > > >>>> project. > > > > > >>>>>> It is released under ALv2 [1] > > > > > >>>>>> > > > > > >>>>>> [1] https://github.com/rustfs/rustfs > > > > > >>>>>> > > > > > >>>>>> Cheers, > > > > > >>>>>> Dmitri. > > > > > >>>>>> > > > > > >>>>>> On Mon, Jan 19, 2026 at 2:13 PM Yufei Gu < > > [email protected]> > > > > > >> wrote: > > > > > >>>>>> > > > > > >>>>>>> Given STS is an essential requirement for s3 compatible > > storage > > > > > >>>>>> validation, > > > > > >>>>>>> waiting for Ozone is reasonable to me. RustFS is also a > good > > > > > >>>> candidate if > > > > > >>>>>>> it's faster and supports STS. With that, are we trying to > > test > > > > > >> both > > > > > >>>> Ozone > > > > > >>>>>>> and RustFS as a long term solution? > > > > > >>>>>>> > > > > > >>>>>>> Yufei > > > > > >>>>>>> > > > > > >>>>>>> > > > > > >>>>>>> On Mon, Jan 19, 2026 at 8:14 AM Jean-Baptiste Onofré < > > > > > >>>> [email protected]> > > > > > >>>>>>> wrote: > > > > > >>>>>>> > > > > > >>>>>>>> Hi Dmitri, > > > > > >>>>>>>> > > > > > >>>>>>>> That’s a fair point. Since we are discussing the > quickstart > > > > > >> guide, > > > > > >>>>>>> ensuring > > > > > >>>>>>>> a consistent message and a positive user experience is > > > > > >> essential. > > > > > >>>>>>>> > > > > > >>>>>>>> While I don't have a strong preference, I lean toward > Apache > > > > > >> Ozone > > > > > >>>> as > > > > > >>>>>> it > > > > > >>>>>>> is > > > > > >>>>>>>> an ASF project, which would help us avoid a similar > > situation > > > > > >> to > > > > > >>>> the > > > > > >>>>>> one > > > > > >>>>>>> we > > > > > >>>>>>>> are currently facing with MinIO. > > > > > >>>>>>>> > > > > > >>>>>>>> That said, if RustFS is faster to set up and works well > with > > > > > >>>> minimal > > > > > >>>>>>>> effort, I am completely fine with that approach. > > > > > >>>>>>>> > > > > > >>>>>>>> Thanks! > > > > > >>>>>>>> > > > > > >>>>>>>> Regards, > > > > > >>>>>>>> JB > > > > > >>>>>>>> > > > > > >>>>>>>> On Mon, Jan 19, 2026 at 3:59 PM Dmitri Bourlatchkov < > > > > > >>>> [email protected]> > > > > > >>>>>>>> wrote: > > > > > >>>>>>>> > > > > > >>>>>>>>> Hi All, > > > > > >>>>>>>>> > > > > > >>>>>>>>> The latest MinIO images work well for Polaris tests and > > > > > >> demos. > > > > > >>>>>>>>> > > > > > >>>>>>>>> It might still be worth proactively trying newer S3 > > > > > >>>> implementations > > > > > >>>>>>> like > > > > > >>>>>>>>> RustFS and/or SeaweedFS (although I have not personally > > tried > > > > > >>>> them > > > > > >>>>>> out > > > > > >>>>>>>> with > > > > > >>>>>>>>> Polaris yet). > > > > > >>>>>>>>> > > > > > >>>>>>>>> Even though we may prefer Ozone as the main storage in > > > > > >> examples > > > > > >>>> when > > > > > >>>>>> it > > > > > >>>>>>>>> supports STS, validating other S3 implementations could > be > > > > > >>>> helpful > > > > > >>>>>> for > > > > > >>>>>>>>> Polaris end users. > > > > > >>>>>>>>> > > > > > >>>>>>>>> Also, I expect RustFS to be faster and less cumbersome to > > > > > >> set up > > > > > >>>> than > > > > > >>>>>>>>> Ozone, which could be an advantage for integration tests > in > > > > > >> CI. > > > > > >>>>>>>>> > > > > > >>>>>>>>> Cheers, > > > > > >>>>>>>>> Dmitri. > > > > > >>>>>>>>> > > > > > >>>>>>>>> On Mon, Jan 19, 2026 at 6:03 AM Jean-Baptiste Onofré < > > > > > >>>>>> [email protected]> > > > > > >>>>>>>>> wrote: > > > > > >>>>>>>>> > > > > > >>>>>>>>>> Hi, > > > > > >>>>>>>>>> > > > > > >>>>>>>>>> I know the Ozone community is discussing to have STS > full > > > > > >>>> support > > > > > >>>>>> for > > > > > >>>>>>>>> 2.2.0 > > > > > >>>>>>>>>> release. > > > > > >>>>>>>>>> > > > > > >>>>>>>>>> So, I think it's reasonable to keep minio for now > waiting > > > > > >> ozone > > > > > >>>>>>> 2.2.0. > > > > > >>>>>>>>>> > > > > > >>>>>>>>>> Thoughts ? > > > > > >>>>>>>>>> > > > > > >>>>>>>>>> Regards > > > > > >>>>>>>>>> JB > > > > > >>>>>>>>>> > > > > > >>>>>>>>>> On Mon, Jan 19, 2026 at 3:59 AM Yong Zheng < > > > > > >> [email protected]> > > > > > >>>>>>> wrote: > > > > > >>>>>>>>>> > > > > > >>>>>>>>>>> Hi Adnan and JB, > > > > > >>>>>>>>>>> > > > > > >>>>>>>>>>> I took a closer look at Ozone and it doesn't appear STS > > > > > >> is > > > > > >>>> fully > > > > > >>>>>>>>>> supported > > > > > >>>>>>>>>>> yet for Ozone: > > > > > >>>>>>>>>>> > > > > > >>>>>>>>>> > > > > > >>>>>>>>> > > > > > >>>>>>>> > > > > > >>>>>>> > > > > > >>>>>> > > > > > >>>> > > > > > >> > > > > > > > > > > > > https://github.com/apache/ozone/blob/master/hadoop-hdds/docs/content/design/ozone-sts.md > > > > > >>>>>>>>>> . > > > > > >>>>>>>>>>> In this case, do we still wan to cut over the getting > > > > > >> start > > > > > >>>>>> example > > > > > >>>>>>>> to > > > > > >>>>>>>>>>> Ozone one instead of MinIO? > > > > > >>>>>>>>>>> > > > > > >>>>>>>>>>> Thanks, > > > > > >>>>>>>>>>> Yong Zheng > > > > > >>>>>>>>>>> > > > > > >>>>>>>>>>> On 2026/01/14 23:32:11 Adnan Hemani via dev wrote: > > > > > >>>>>>>>>>>> Hi all, > > > > > >>>>>>>>>>>> > > > > > >>>>>>>>>>>> I'd hope that we can pivot to Ozone and hide the > > > > > >> additional > > > > > >>>>>>>>> adaptation > > > > > >>>>>>>>>>>> behind our setup scripts. > > > > > >>>>>>>>>>>> > > > > > >>>>>>>>>>>> It's a shame that my fears came true about relying on > a > > > > > >>>>>> provider > > > > > >>>>>>>> who > > > > > >>>>>>>>> is > > > > > >>>>>>>>>>>> actively trying to sell the managed version of their > > > > > >>>> product > > > > > >>>>>>>> shutting > > > > > >>>>>>>>>>> down > > > > > >>>>>>>>>>>> their OSS functionalities [1]. Any alternatives we > > > > > >>>> consider, in > > > > > >>>>>>> my > > > > > >>>>>>>>>>> opinion, > > > > > >>>>>>>>>>>> should take this experience into consideration. > > > > > >>>>>>>>>>>> > > > > > >>>>>>>>>>>> I'm still not against reverting back to using local FS > > > > > >> for > > > > > >>>> the > > > > > >>>>>>>>>> Quickstart > > > > > >>>>>>>>>>>> experience. If all the alternatives only complicate > > > > > >> things > > > > > >>>> for > > > > > >>>>>>> the > > > > > >>>>>>>>> end > > > > > >>>>>>>>>>>> user, then it will be a poor Quickstart experience - > > > > > >> and > > > > > >>>> that's > > > > > >>>>>>>> much > > > > > >>>>>>>>>>> worse > > > > > >>>>>>>>>>>> than giving more functionality in Quickstart but users > > > > > >> not > > > > > >>>>>>> knowing > > > > > >>>>>>>>> how > > > > > >>>>>>>>>> to > > > > > >>>>>>>>>>>> actually use it. > > > > > >>>>>>>>>>>> > > > > > >>>>>>>>>>>> Best, > > > > > >>>>>>>>>>>> Adnan Hemani > > > > > >>>>>>>>>>>> > > > > > >>>>>>>>>>>> [1] https://github.com/apache/polaris/pull/2976 > > > > > >>>>>>>>>>>> > > > > > >>>>>>>>>>>> On Fri, Jan 9, 2026 at 6:58 AM Jean-Baptiste Onofré < > > > > > >>>>>>>> [email protected] > > > > > >>>>>>>>>> > > > > > >>>>>>>>>>> wrote: > > > > > >>>>>>>>>>>> > > > > > >>>>>>>>>>>>> Hi > > > > > >>>>>>>>>>>>> > > > > > >>>>>>>>>>>>> Yes, Ozone supports STS, that's why I mentioned it. > > > > > >> It's > > > > > >>>>>>> probably > > > > > >>>>>>>>>> close > > > > > >>>>>>>>>>>>> enough to minIO (even if some "adaptation" might be > > > > > >>>> required > > > > > >>>>>>> for > > > > > >>>>>>>>>> STS). > > > > > >>>>>>>>>>>>> > > > > > >>>>>>>>>>>>> Just my $0.01 > > > > > >>>>>>>>>>>>> > > > > > >>>>>>>>>>>>> Regards > > > > > >>>>>>>>>>>>> JB > > > > > >>>>>>>>>>>>> > > > > > >>>>>>>>>>>>> On Fri, Jan 9, 2026 at 2:15 AM Yufei Gu < > > > > > >>>>>> [email protected]> > > > > > >>>>>>>>>> wrote: > > > > > >>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>> Agreed with Dmitri that STS is essential in this > > > > > >> case, > > > > > >>>>>>>> otherwise > > > > > >>>>>>>>> a > > > > > >>>>>>>>>>> local > > > > > >>>>>>>>>>>>>> file system will be good enough. > > > > > >>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>> AFAIK, Apache Ozone supports STS starting from > > > > > >> 2.1.0, > > > > > >>>>>>>>>>>>>> https://ozone.apache.org/release/2.1.0/. > > > > > >>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>> Yufei > > > > > >>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>> On Thu, Jan 8, 2026 at 4:55 PM Dmitri Bourlatchkov > > > > > >> < > > > > > >>>>>>>>>> [email protected] > > > > > >>>>>>>>>>>> > > > > > >>>>>>>>>>>>>> wrote: > > > > > >>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>> Hi François, > > > > > >>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>> I could not find any mention of STS / AssumeRole > > > > > >> in > > > > > >>>>>>> GarageHQ > > > > > >>>>>>>>> docs > > > > > >>>>>>>>>>>>> (really > > > > > >>>>>>>>>>>>>>> quick looks), so I assume it does not support > > > > > >> STS... > > > > > >>>> Do > > > > > >>>>>> you > > > > > >>>>>>>>> know > > > > > >>>>>>>>>>> for > > > > > >>>>>>>>>>>>>> sure? > > > > > >>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>> STS is pretty important for ease of use in > > > > > >> Polaris > > > > > >>>>>> getting > > > > > >>>>>>>>>> started > > > > > >>>>>>>>>>>>> (i.e. > > > > > >>>>>>>>>>>>>>> setup without STS is more involved). > > > > > >>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>> Cheers, > > > > > >>>>>>>>>>>>>>> Dmitri. > > > > > >>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>> On Thu, Jan 8, 2026 at 3:09 PM Francois Papon < > > > > > >>>>>>>>> [email protected] > > > > > >>>>>>>>>>> > > > > > >>>>>>>>>>>>> wrote: > > > > > >>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>> Hi, > > > > > >>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>> About other solutions, I made some test to > > > > > >> evaluate > > > > > >>>>>>> GaragHQ > > > > > >>>>>>>>> as > > > > > >>>>>>>>>> an > > > > > >>>>>>>>>>>>>>>> alternative to MinIO: > > > > > >>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>> https://garagehq.deuxfleurs.fr/ > > > > > >>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>> I don't know if all the required > > > > > >> functionnalities > > > > > >>>> used > > > > > >>>>>> by > > > > > >>>>>>>>>>> Polaris are > > > > > >>>>>>>>>>>>>>>> provided but for the test/quickstart part may > > > > > >> be > > > > > >>>> it can > > > > > >>>>>>> do > > > > > >>>>>>>>> the > > > > > >>>>>>>>>>> job. > > > > > >>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>> regards, > > > > > >>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>> François > > > > > >>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>> Le 08/01/2026 à 16:35, Adam Christian a écrit : > > > > > >>>>>>>>>>>>>>>>> Thanks, folks! Great points! I do like the > > > > > >> idea > > > > > >>>> of > > > > > >>>>>>> using > > > > > >>>>>>>>>>> something > > > > > >>>>>>>>>>>>>>> other > > > > > >>>>>>>>>>>>>>>>> than MinIO. > > > > > >>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>> My only 2 cents: For the quickstart docker > > > > > >>>> compose > > > > > >>>>>>> file, > > > > > >>>>>>>> I > > > > > >>>>>>>>>>> believe > > > > > >>>>>>>>>>>>>> that > > > > > >>>>>>>>>>>>>>>> we > > > > > >>>>>>>>>>>>>>>>> should use object storage that can be > > > > > >>>> automatically > > > > > >>>>>>>>>> configured > > > > > >>>>>>>>>>>>>> without > > > > > >>>>>>>>>>>>>>>>> having to put in credentials. The goal of the > > > > > >>>>>>> quickstart > > > > > >>>>>>>> is > > > > > >>>>>>>>>> to > > > > > >>>>>>>>>>> have > > > > > >>>>>>>>>>>>>>>> anyone > > > > > >>>>>>>>>>>>>>>>> who has Docker running locally to be able to > > > > > >>>> launch a > > > > > >>>>>>>>>>>>> non-production > > > > > >>>>>>>>>>>>>>>>> version of Polaris through a single command. > > > > > >> It's > > > > > >>>>>>>> targeted > > > > > >>>>>>>>>> for > > > > > >>>>>>>>>>>>> users > > > > > >>>>>>>>>>>>>>> who > > > > > >>>>>>>>>>>>>>>>> are just learning about Polaris. So, in my > > > > > >>>> opinion, > > > > > >>>>>>> this > > > > > >>>>>>>>>> rules > > > > > >>>>>>>>>>> out > > > > > >>>>>>>>>>>>>> AWS > > > > > >>>>>>>>>>>>>>>> S3. > > > > > >>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>> As long as we meet that criteria, I do think > > > > > >> that > > > > > >>>>>>>> changing > > > > > >>>>>>>>>>> would be > > > > > >>>>>>>>>>>>>>>>> appropriate. Thanks for bringing this up! > > > > > >>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>> Go community, > > > > > >>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>> Adam > > > > > >>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>> On Thu, Jan 8, 2026 at 9:16 AM Dmitri > > > > > >>>> Bourlatchkov < > > > > > >>>>>>>>>>>>>>>>> [email protected]> wrote: > > > > > >>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>> Hi All, > > > > > >>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>> Apache Ozone works well, AFAIK [1] but it > > > > > >>>> certainly > > > > > >>>>>>>>> requires > > > > > >>>>>>>>>>> more > > > > > >>>>>>>>>>>>>>> setup > > > > > >>>>>>>>>>>>>>>>>> work than MinIO. > > > > > >>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>> WDYT about RustFS? [2] > > > > > >>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>> [1] > > > > > >>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>> > > > > > >>>>>>>>>>> > > > > > >>>>>>>>>> > > > > > >>>>>>>>> > > > > > >>>>>>>> > > > > > >>>>>>> > > > > > >>>>>> > > > > > >>>> > > > > > >> > > > > > > > > > > > > https://polaris.apache.org/in-dev/unreleased/getting-started/creating-a-catalog/s3/catalog-ozone/ > > > > > >>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>> [2] https://github.com/rustfs/rustfs > > > > > >>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>> Cheers, > > > > > >>>>>>>>>>>>>>>>>> Dmitri. > > > > > >>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>> On Thu, Jan 8, 2026 at 8:46 AM Jean-Baptiste > > > > > >>>> Onofré > > > > > >>>>>> < > > > > > >>>>>>>>>>>>>> [email protected]> > > > > > >>>>>>>>>>>>>>>>>> wrote: > > > > > >>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>>> Hi Yong, > > > > > >>>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>>> Thanks for bringing this to our attention. > > > > > >>>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>>> I agree with your assessment regarding the > > > > > >>>> recent > > > > > >>>>>>>> changes > > > > > >>>>>>>>>> to > > > > > >>>>>>>>>>>>> MinIO > > > > > >>>>>>>>>>>>>>> OSS. > > > > > >>>>>>>>>>>>>>>>>> We > > > > > >>>>>>>>>>>>>>>>>>> could evaluate Apache Ozone as a potential > > > > > >>>>>>> alternative; > > > > > >>>>>>>>>>> while STS > > > > > >>>>>>>>>>>>>>> might > > > > > >>>>>>>>>>>>>>>>>>> require more changes, I believe it’s worth > > > > > >>>>>>>> investigating. > > > > > >>>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>>> In the short term, we could update the > > > > > >>>> quickstart > > > > > >>>>>>> guide > > > > > >>>>>>>>> to > > > > > >>>>>>>>>>> use > > > > > >>>>>>>>>>>>> AWS > > > > > >>>>>>>>>>>>>> S3 > > > > > >>>>>>>>>>>>>>>>>> while > > > > > >>>>>>>>>>>>>>>>>>> we wait to finalize the Ozone integration. > > > > > >>>>>>>>>>>>>>>>>>> About Ceph, that's a good idea, assuming we > > > > > >>>> have > > > > > >>>>>> the > > > > > >>>>>>>> same > > > > > >>>>>>>>>>> kind of > > > > > >>>>>>>>>>>>>>>>>>> layout/experience. > > > > > >>>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>>> Regards, > > > > > >>>>>>>>>>>>>>>>>>> JB > > > > > >>>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>>> On Thu, Jan 8, 2026 at 7:49 AM Yong Zheng < > > > > > >>>>>>>>>> [email protected] > > > > > >>>>>>>>>>>> > > > > > >>>>>>>>>>>>>> wrote: > > > > > >>>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>>>> Hello, > > > > > >>>>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>>>> We have MinIO support as S3 compatible > > > > > >>>> storage and > > > > > >>>>>>>> this > > > > > >>>>>>>>> is > > > > > >>>>>>>>>>> great > > > > > >>>>>>>>>>>>>> as > > > > > >>>>>>>>>>>>>>> it > > > > > >>>>>>>>>>>>>>>>>>>> allows users to quickly test out Apache > > > > > >>>> Polaris > > > > > >>>>>> as a > > > > > >>>>>>>>>>> catalog and > > > > > >>>>>>>>>>>>>>> write > > > > > >>>>>>>>>>>>>>>>>>> to a > > > > > >>>>>>>>>>>>>>>>>>>> S3 compatible storage. However, as MinIO > > > > > >> is > > > > > >>>> now > > > > > >>>>>>> under > > > > > >>>>>>>>>>>>> maintenance > > > > > >>>>>>>>>>>>>>> mode > > > > > >>>>>>>>>>>>>>>>>>> only > > > > > >>>>>>>>>>>>>>>>>>>> for OSS ( > > > > > >>>>>>>>>> https://github.com/minio/minio?tab=readme-ov-file), > > > > > >>>>>>>>>>> we > > > > > >>>>>>>>>>>>>>> won't > > > > > >>>>>>>>>>>>>>>>>> be > > > > > >>>>>>>>>>>>>>>>>>>> able to get updated images from public > > > > > >> image > > > > > >>>>>>> registry, > > > > > >>>>>>>>>>> should we > > > > > >>>>>>>>>>>>>>>>>> consider > > > > > >>>>>>>>>>>>>>>>>>>> switch our primary getting-start example > > > > > >> to > > > > > >>>>>>> non-MinIO > > > > > >>>>>>>>> one > > > > > >>>>>>>>>>>>> instead > > > > > >>>>>>>>>>>>>>> (the > > > > > >>>>>>>>>>>>>>>>>>>> current one is MinIO as backend: > > > > > >>>>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>> > > > > > >>>>>>>>>>> > > > > > >>>>>>>>>> > > > > > >>>>>>>>> > > > > > >>>>>>>> > > > > > >>>>>>> > > > > > >>>>>> > > > > > >>>> > > > > > >> > > > > > > > > > > > > https://github.com/apache/polaris/blob/main/getting-started/quickstart/docker-compose.yml > > > > > >>>>>>>>>>>>>>>>>>> )? > > > > > >>>>>>>>>>>>>>>>>>>> Without doing so, users will be likely > > > > > >> pulling > > > > > >>>>>> down > > > > > >>>>>>>>>> outdated > > > > > >>>>>>>>>>>>> MinIO > > > > > >>>>>>>>>>>>>>>>>> images > > > > > >>>>>>>>>>>>>>>>>>>> with critical CVEs couple months later for > > > > > >>>> their > > > > > >>>>>>> local > > > > > >>>>>>>>>>> setup to > > > > > >>>>>>>>>>>>>> play > > > > > >>>>>>>>>>>>>>>>>>>> around. If using outdated MinIO is a > > > > > >> concern > > > > > >>>> as > > > > > >>>>>> the > > > > > >>>>>>>>>>>>> getting-start > > > > > >>>>>>>>>>>>>>>>>>> example, > > > > > >>>>>>>>>>>>>>>>>>>> maybe we should switch to the Ceph one ( > > > > > >>>>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>> > > > > > >>>>>>>>>>> > > > > > >>>>>>>>>> > > > > > >>>>>>>>> > > > > > >>>>>>>> > > > > > >>>>>>> > > > > > >>>>>> > > > > > >>>> > > > > > >> > > > > > > > > > > > > https://github.com/apache/polaris/blob/main/getting-started/ceph/docker-compose.yml > > > > > >>>>>>>>>>>>>>>>>>> ) > > > > > >>>>>>>>>>>>>>>>>>>> but updated it to match the same layout? > > > > > >>>>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>>>>>> Thanks, > > > > > >>>>>>>>>>>>>>>>>>>> Yong Zheng > > > > > >>>>>>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>>> > > > > > >>>>>>>>>>>>> > > > > > >>>>>>>>>>>> > > > > > >>>>>>>>>>> > > > > > >>>>>>>>>> > > > > > >>>>>>>>> > > > > > >>>>>>>> > > > > > >>>>>>> > > > > > >>>>>> > > > > > >>>>> > > > > > >>>> > > > > > >>> > > > > > >> > > > > > > > > > > > >
