[
https://issues.apache.org/jira/browse/HDDS-7594?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17651750#comment-17651750
]
Neil Joshi commented on HDDS-7594:
----------------------------------
Thanks [~erose] for your comments and feedback on the s3g fso layout
interoperability with ofs when dealing with folders in the path. Currently
when applications attempt to create folders over the s3g with fso layout, they
result in zero byte files. This occurs when put requests are issued over s3g
with applications expecting folder creation prior to placing keys in the folder
path, the
{quote}{{put key1}}{{ followed by }}{{put key1/key2}}{{ when running on S3? }}
{quote}
We are expecting this to occur when using applications such as trino
interfacing with ozone through both ofs and s3g with the fso bucket layout. We
have also encountered the problem in an example trino setup to create tables
over s3g. When using legacy bucket layout, the example successfully creates
and populates the table. However, with same example using the FSO layout we've
see the following error from a thrown exception,
*Got exception: org.apache.hadoop.fs.FileAlreadyExistsException Path is a file:
s3a://fso/customer*
where _customer_ is a directory created under bucket _fso_ for table _customer_
that should be populated with key entries. I'll follow up with more detail on
the errors encountered if I can.
I agree that,
{quote}I think we would need a client side config (for ozone client on s3
gateway) to convert 0 byte file writes from S3 API to directory creates if the
underlying bucket type is FSO.
{quote}
We should look to implement something like this to resolve this issue. Let me
know what you would like to see included in the client side implementation for
this issue.
> [FSO] Folders created through S3G are created on file system as "files".
> ------------------------------------------------------------------------
>
> Key: HDDS-7594
> URL: https://issues.apache.org/jira/browse/HDDS-7594
> Project: Apache Ozone
> Issue Type: Bug
> Reporter: Neil Joshi
> Priority: Major
>
> With FSO bucket layout buckets, folders created with the s3-gateway create 0
> byte files on file system. Unlike Object Store bucket layout buckets that
> create folders correctly when created through the s3-gateway, folder creation
> on FSO buckets with s3 gateway results in 0 byte files. Expected is the same
> behaviour. This has caused problems with interoperability for applications
> accessing ozone through both ofs and the s3 gateway, such as Trino.
>
> Problem can be produced issuing aws-cli folder creation to the s3 gateway on
> an ozone cluster layout {*}FILE_SYSTEM_OPTIMIZED{*}:
>
> {code:java}
> bash-4.2$ ozone sh bucket create --layout=FILE_SYSTEM_OPTIMIZED /s3v/bucketfso
> ozone$ aws s3api --profile ozone --endpoint http://localhost:9878 put-object
> --bucket bucketfso --key test1
> ozone$ aws s3api --profile ozone --endpoint http://localhost:9878 put-object
> --bucket bucketfso --key test1/README.md --body ./README.md
> An error occurred (InvalidRequest) when calling the PutObject operation: An
> error occurred (InvalidRequest) when calling the PutObject/MPU PartUpload
> operation: ozone.om.enable.filesystem.paths is enabled Keys are considered as
> Unix Paths. Path has Violated FS Semantics which caused put operation to fail.
> bash-4.2$ ozone fs -ls ofs://om/s3v/bucketfso Found 2 items -rw-rw-rw- 1
> hadoop hadoop 0 2022-12-05 22:34 ofs://om/s3v/bucketfso/test1
> {code}
> File is created _when_ a folder is expected. If it were correctly set it
> would result in seeing _*drwxrwxrwx*_ for the test1 key as in,
> {code:java}
> drwxrwxrwx - hadoop hadoop 0 2022-12-05 22:40
> ofs://om/s3v/bucketfos/test1
> {code}
>
> In contrast, the same aws command issued on a *OBJECT_STORE* layout bucket
> results in :
> {code:java}
> bash-4.2$ ozone sh key list /s3v/bucketobs
> [ {
> "volumeName" : "s3v",
> "bucketName" : "bucketobs",
> "name" : "test1",
> "dataSize" : 0,
> "creationTime" : "2022-12-05T22:33:21.737Z",
> "modificationTime" : "2022-12-05T22:33:21.746Z",
> "replicationConfig" : {
> "replicationFactor" : "ONE",
> "requiredNodes" : 1,
> "replicationType" : "RATIS"
> },
> "metadata" : { }
> }, {
> "volumeName" : "s3v",
> "bucketName" : "bucketobs",
> "name" : "test1/README.md",
> "dataSize" : 3879,
> "creationTime" : "2022-12-05T22:34:02.402Z",
> "modificationTime" : "2022-12-05T22:37:00.758Z",
> "replicationConfig" : {
> "replicationFactor" : "ONE",
> "requiredNodes" : 1,
> "replicationType" : "RATIS"
> },
> "metadata" : { }
> } ]{code}
> As is expected.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
