NFS WORM (Write Once, Read Many)

In detail, creating files through NFS is a multi step process. To write to a new file, NFS client first sends the CREATE request with no payload to NFS server. After receiving a response, the server issues a WRITE request. It is a problem for FS enabled buckets under retention as the file created with 0 bytes blocks any writes to it. Due to this reason, until ECS 3.3, retention on FS enabled bucket makes the whole mounted file-system read-only. There is no End of File (EOF) concept in NFS. Thus, setting a retention for files, on the FS enabled buckets, after writing to them does not work as expected.

To remove the constraints that are placed on NFS files in a retention enabled bucket, the autocommit period is implemented on NFS data. For this reason, it is decided to introduce the autocommit period during which certain types of updates (for now identified as writes, Acl updates and deletes that are required for rsync, and rename that is required for Vim editor) are allowed, which removes the retention constraints for that period alone.

NOTE:  The autocommit and the Atmos retention start delay are the same. Autocommit period is a bucket property like retention period. Autocommit period is, Applicable only for the file system enabled buckets with retention period Applicable to the buckets in noncompliant namespace Applies to only requests from NFS and Atmos

Seal file

The seal file functionality helps to commit the file to WORM state when the file is written, ignoring the remaining autocommit period. The seal function is performed through the command: chmod ugo-w <file> on the file.

NOTE:  The seal functionality does not have any effect outside the retention period.

High level overview

The following diagram provides an overview of the autocommit period behavior:

Autocommit configuration

The autocommit period can be set from the user interface or bucket REST API or S3 head or Atmos subtenant API.

User Interface

• When the File System is not enabled, no autocommit option is displayed.
• When the File System is enabled /no retention value is specified, autocommit is displayed but disabled.
• When the File System is enabled/retention value selected/autocommit is displayed and enabled for selection.

NOTE: Maximum autocommit period is limited to the smaller of the Bucket Retention period or the default maximum period of one day.

REST API

lglou063:~ # curl -i -k -T /tmp/bucket -X POST https://10.247.99.11:4443/object/bucket -H "$token" -H "Content-Type: application/xml" -v

The contents of /tmp/bucket
<object_bucket_create>
	<name>bucket2</name>
	<namespace>s3</namespace>
	<filesystem_enabled>true</filesystem_enabled>
	<autocommit_period>300</autocommit_period>
	<retention>1500</retention>
</object_bucket_create>

S3 head

Bucket creation

Bucket creation flow through s3 head can make use of optional request header, x-emc-auto-commit-period:seconds to set the autocommit period. The following checks are made in this flow:

• Allow only positive integers
• Settable only for file system buckets
• Settable only when the retention value is present

./s3curl.pl --ord --id=naveen --key=+1Zh4YC2r2puuUaj3Lbnj3u0G9qgPRj0RIWJhPxH --createbucket -- -H 'x-emc-autocommit-period:600' -H 'x-emc-file-system-access-enabled:true' -H 'x-emc-namespace:ns1' http://10.249.245.187:9020/bucket5 -v

Atmos

Atmos creates a subtenant request header, x-emc-retention-start-delay, captures the autocommit interval.

./atmoscurl.pl -user USER1 -action PUT -pmode TID -path / -header "x-emc-retention-period:300" -header "x-emc-retention-start-delay:120" -include

Behavior of file operations