Lewis John McGibbney created NUTCH-3244:
-------------------------------------------
Summary: Rewrite the Nutch metrics reference documentation
Key: NUTCH-3244
URL: https://issues.apache.org/jira/browse/NUTCH-3244
Project: Nutch
Issue Type: Sub-task
Reporter: Lewis John McGibbney
[NUTCH-2909|https://issues.apache.org/jira/browse/NUTCH-2909] created the
[Nutch wiki Metrics
page|https://cwiki.apache.org/confluence/display/NUTCH/Metrics] and its
[metrics
table|https://cwiki.apache.org/confluence/display/NUTCH/Metrics#Metrics-MetricsTable].
Both still use the counter names from before
[NUTCH-3132|https://issues.apache.org/jira/browse/NUTCH-3132]. The metrics
report's suggestion 11 asks for a reference entry per metric, an expected
range, and a suggested alert threshold. None of those are on the wiki.
Most constants in \{{NutchMetrics}} have a single-line javadoc saying what the
counter counts. The latency prefixes (\{{FETCHER_LATENCY}},
\{{PARSER_LATENCY}}, \{{INDEXER_LATENCY}}) and the \{{ERROR_*}} constants have
longer javadoc. None of them state a range or an alert threshold. The \{{_ms}}
suffixes an operator sees (\{{fetch_latency_p50_ms}} and the rest) are built in
\{{LatencyTracker}}, not declared as constants in \{{NutchMetrics}}.
h2. Proposal
* Rewrite the wiki Metrics page and the metrics table so every row uses the
current group and counter name, including the latency names \{{LatencyTracker}}
actually emits (\{{_count_total}}, \{{_sum_ms}}, \{{_p50_ms}}, \{{_p95_ms}},
\{{_p99_ms}}). Keep the old names out of this table. Link to the migration
guide from issue 2.
* Add one row per counter: group, name, the component that emits it, what it
counts, the unit, a plausible range for a healthy crawl, and a suggested alert.
Mark ranges and thresholds as operator guidance, since they depend on crawl
size and configuration.
* Where a constant's javadoc no longer matches the counter it names, correct
the javadoc. Do not copy the ranges and thresholds into every constant.
h2. Acceptance
* Every counter name emitted by \{{NutchMetrics}} or \{{LatencyTracker}}
appears in the wiki table with its current name.
* No row uses a pre-NUTCH-3132 name.
* The page links to the migration guide for the old names.
* Each row has a unit, a range, and a suggested alert, or an explicit "n/a"
where a range does not make sense (for example a gauge published with
\{{setValue}}).
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
