HDDS-11946. Require all ozone repair commands to support a --dry-run option

[adoroszlai]

What changes were proposed in this pull request?

1. Add --dry-run option to all ozone repair subcommands that modify state. The new option currently defaults to false, so commands perform repair by default. (Otherwise we'd need to use it as --dry-run=false.)
2. Implement dry-run mode where needed.

https://issues.apache.org/jira/browse/HDDS-11946

How was this patch tested?

Enforce --dry-run option

Added unit test to verify that each subcommand under ozone repair satisfies one of the following:

• has subcommands
• has --dry-run option
• implements ReadOnlySubcommand (currently only QuotaStatus)

Test failure before implementing ReadOnlyCommand in QuotaStatus:

[ERROR] Tests run: 3, Failures: 1, Errors: 0, Skipped: 0, Time elapsed: 0.261 s <<< FAILURE! - in org.apache.hadoop.ozone.repair.TestOzoneRepair
[ERROR] org.apache.hadoop.ozone.repair.TestOzoneRepair.subcommandsSupportDryRun  Time elapsed: 0.211 s  <<< FAILURE!
java.lang.AssertionError: 
['ozone repair quota status' defined by class org.apache.hadoop.ozone.repair.quota.QuotaStatus should support --dry-run or implement interface org.apache.hadoop.ozone.repair.ReadOnlyCommand] 
Expecting UnmodifiableSet:
  ["--service-id",
    "--om-service-id",
    "--service-host",
    "-h",
    "--help",
    "-V",
    "--version"]
to contain:
  ["--dry-run"]
but could not find the following element(s):
  ["--dry-run"]

	at org.apache.hadoop.ozone.repair.TestOzoneRepair.assertSubcommandOptionRecursively(TestOzoneRepair.java:84)
	at org.apache.hadoop.ozone.repair.TestOzoneRepair.assertSubcommandOptionRecursively(TestOzoneRepair.java:89)
	at org.apache.hadoop.ozone.repair.TestOzoneRepair.assertSubcommandOptionRecursively(TestOzoneRepair.java:89)
	at org.apache.hadoop.ozone.repair.TestOzoneRepair.subcommandsSupportDryRun(TestOzoneRepair.java:71)

Tool operation

Sample output of ozone repair om fso-tree from integration tests:

[dry run] No running OM service detected. Proceeding with repair.
[dry run] Creating database of reachable directories at hadoop-ozone/integration-test/target/test-dir/MiniOzoneClusterImpl-430e6515-4d93-4d63-99d9-2d0e1801aa90/ozone-meta/reachable.db
[dry run] Processing volume: /vol2
[dry run] Processing bucket: vol2/bucket1
[dry run] Deleting unreferenced directory /-9223372036854768896/-9223372036854768384/-9223372036854767871/dir2
[dry run] Deleting unreferenced file /-9223372036854768896/-9223372036854768384/-9223372036854767359/file3
[dry run] Deleting unreferenced file /-9223372036854768896/-9223372036854768384/-9223372036854767871/file1
[dry run] Deleting unreferenced file /-9223372036854768896/-9223372036854768384/-9223372036854767871/file2
[dry run] Processing bucket: vol2/bucket2
2025-01-10 16:46:50,087 [main] INFO  rocksdiff.RocksDBCheckpointDiffer (RocksDBCheckpointDiffer.java:close(307)) - Shutting down CompactionDagPruningService.
[dry run] 
Reachable:
	Directories: 4
	Files: 5
	Bytes: 50
Unreachable:
	Directories: 0
	Files: 0
	Bytes: 0
Unreferenced:
	Directories: 1
	Files: 3
	Bytes: 30

Manual test of ozone repair om update-transaction in compose environment:

$ ozone repair om update-transaction --db=/data/metadata/om.db --term 2 --index 3 --dry-run
[dry run] No running OM service detected. Proceeding with repair.
[dry run] The original highest transaction Info was (t:1, i:14)
[dry run] Updating transaction info to (t:2, i:3)

$ ozone repair om update-transaction --db=/data/metadata/om.db --term 2 --index 3 
ATTENTION: Running as user hadoop. Make sure this is the same user used to run the Ozone process. Are you sure you want to continue (y/N)? y
Run as user: hadoop
No running OM service detected. Proceeding with repair.
The original highest transaction Info was (t:1, i:14)
Updating transaction info to (t:2, i:3)
The highest transaction info has been updated to: (t:2, i:3)

$ ozone repair om update-transaction --db=/data/metadata/om.db --term 2 --index 3 --dry-run
[dry run] No running OM service detected. Proceeding with repair.
[dry run] The original highest transaction Info was (t:2, i:3)
[dry run] Updating transaction info to (t:2, i:3)

CI:
https://github.com/adoroszlai/ozone/actions/runs/12714010228

[adoroszlai]

@sarvekshayr @Tejaskriya please review

[sarvekshayr]

In my opinion, it would be better to make the --dry-run option the default behaviour. This allows users to see what changes would be performed before executing the actual repair, reducing the risk of accidental modifications to the database. While users can explicitly specify the --dry-run option to preview changes, making it the default ensures it won't be overlooked.

Since the primary focus of this PR is to add the --dry-run option for all repair commands (and any commands to be added in the future), we can consider making --dry-run the default behaviour in a follow-up JIRA.

[adoroszlai]

we can consider making --dry-run the default behaviour in a follow-up JIRA.

I think we can decide and implement the default behavior here, that's why I initially created the PR as draft.

I have no objection against making dry run the default, but I find --dry-run=false cumbersome. IMO we need a different name for the option in this case. --repair is already used in fso-tree, so that would be my primary choice.

[sarvekshayr]

I have no objection against making dry run the default, but I find --dry-run=false cumbersome. IMO we need a different name for the option in this case. --repair is already used in fso-tree, so that would be my primary choice.

We can go ahead and use the --repair flag.

[adoroszlai]

@errose28 could you please review?

[errose28]

I actually prefer --dry-run as an option and the default to be doing the repair. The user has already entered ozone repair and acknowledged the warning prompt, so after all this not doing the repair seems counter-intuitive. Not to mention the repetitiveness of ozone repair ... --repair just to make the command do what was intended. FSO repair was originally made before there as a dedicated ozone repair command with its own safeguards in place hence dry run was the default in that implementation, but we have a better framework in place now.