Chi-Hsuan Huang created HDDS-16375:
--------------------------------------
Summary: Recon fileCount API silently ignores the fileSize
parameter unless volume and bucket are both set
Key: HDDS-16375
URL: https://issues.apache.org/jira/browse/HDDS-16375
Project: Apache Ozone
Issue Type: Bug
Components: Ozone Recon
Reporter: Chi-Hsuan Huang
{{GET /api/v1/utilization/fileCount}} applies the {{fileSize}} filter only when
{{volume}}, {{bucket}} and {{fileSize}} are all provided. Any other combination
falls into the scan branch, which filters on volume and bucket only, so
{{fileSize}} is dropped without an error or warning.
Measured on current master by calling {{UtilizationEndpoint#getFileCounts}}
directly from {{TestEndpoints#testGetFileCounts}}. The dataset has 3 bins:
{{vol1/bucket1:1024}}, {{vol1/bucket1:131072}}, {{vol2/bucket1:1024}}.
{noformat}
getFileCounts\(null, null, 131072\) \-> 3 results: \[vol1/bucket1:1024,
vol1/bucket1:131072, vol2/bucket1:1024\]
getFileCounts\("vol1", null, 131072\) \-> 2 results: \[vol1/bucket1:1024,
vol1/bucket1:131072\]
{noformat}
Both should return a single bin. The responses include bins that do not match
the requested size, so the caller receives an unfiltered list that looks like a
filtered one.
ReconApi.md lists {{fileSize}} as optional and describes it as "Filters the
results based on the given fileSize", with no stated dependency on the other
two parameters.
This matters more now that the Recon chatbot \(HDDS\-14816, HDDS\-15570\)
registers this endpoint as an LLM tool and advertises {{fileSize}} as a
parameter in {{LlmToolSpecFactory}}. A model answering "how many files around
1MB are there" will pass {{fileSize}} on its own and present the full
distribution as a filtered answer.
Two ways to resolve this, and the choice needs a maintainer call:h1. Apply the
{{fileSize}} filter independently in the scan branch, matching the documented
behaviour.
h1. Keep the current parameter hierarchy and update ReconApi.md to state that
{{fileSize}} requires {{volume}} and {{bucket}}.
This is pre\-existing and dates back to HDDS\-3681. HDDS\-16357 fixed a
separate normalization bug in the same method and does not overlap with this.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
