HDDS-16965. Initialize disk balancer state from configured values when diskBalancer.info is absent

[shuan1026]

What changes were proposed in this pull request?

Bug: On first DiskBalancer init (diskBalancer.info absent), only should.run.default was taken from injected config; threshold, bandwidth, parallel thread, stop-after-disk-even, and container states were seeded via new DiskBalancerConfiguration() and persisted as Java defaults. Later restarts keep the wrong file; XML changes alone do not fix it.

Impact: Operator ozone-site.xml tuning is ignored on new datanodes. Balancing may use wrong I/O limits / stop condition / movable states; diskbalancer status reports the persisted defaults with no warning.

Fix: Seed initial DiskBalancerInfo from conf.getObject(DiskBalancerConfiguration.class). Existing-file load path unchanged.

What is the link to the Apache JIRA

https://issues.apache.org/jira/browse/HDDS-16965

How was this patch tested?

• Added TestDiskBalancerService#testInitialDiskBalancerInfoUsesConfiguredValues: empty info dir + non-default config via setFromObject; asserts in-memory info and written YAML match configured values.
• ci: https://github.com/shuan1026/ozone/actions/runs/38335830063

[Gargi-jais11]

Thanks @shuan1026 for finding out this issue.
Please find the inlined comment.

[Gargi-jais11]

LGTM, +1!

[slfan1989]

Thanks for inviting me to review this PR.

I left one inline comment about an edge case in threshold validation.

[slfan1989]

NaN bypasses this validation because both comparisons evaluate to false.

Since the configuration parser uses Double.parseDouble(), an XML value of NaN reaches this method and is subsequently persisted to diskBalancer.info.

Once balancing runs, comparisons against the resulting NaN threshold are always false. For example, newUtilization >= upperThreshold cannot reject a move, which may cause containers to be selected without a valid utilization limit.

Could we reject non-finite values here and in setThreshold(), for example:

if (!Double.isFinite(threshold) || threshold <= 0d || threshold >= 100d)

Please also add a test using a raw NaN configuration value.

[shuan1026]

Thanks for the review @slfan1989!
Done, NaN is now rejected in both validate() and setThreshold(), with tests added.

[slfan1989]

Thanks for addressing the feedback.

The non-finite threshold validation is now shared by configuration initialization and the setter, and the tests cover NaN and infinities.

LGTM, +1!