[#1135] Head the last near-limit column of backendstat show-index-status 80%, the threshold it counts from

[vharseko]

Fixes #1135

Problem

The last column of backendstat show-index-status was headed 85%, but it counted the keys holding from 80% up to 90% of the index entry limit: nearLimit let a key in from entryLimit * 0.8, and every near key below 0.9 fell into that column.

Fix

The heading now reads 80%, the threshold the column actually counts from. The threshold is kept: the counter was named eighty, the existing test pins "from 80% on", and moving the bucket to 0.85 would silently drop the keys between 80% and 85% of the limit from the report.

So that the heading and the threshold cannot disagree again, both come from one list:

• NEAR_LIMIT_PERCENTS = { 95, 90, 80 } gives the headings and the thresholds;
• nearLimit() becomes nearLimitColumn(), which returns the column counting a key, or -1 when the key is not near its limit (an index-entry-limit of 0 still means no key is near it, An index-entry-limit of 0 is treated as the smallest limit when a change of the limit decides whether the index must be rebuilt #1059). The comparison is exact integer arithmetic (size * 100 >= entryLimit * percent) instead of entryLimit * 0.95 in double;
• the three counters become one array, appended in the order of the headings.

The sample output in the Administration Guide (Indexing) shows the 80% heading.

Tests

BackendStatTest:

• the two existing cases now go through nearLimitColumn();
• testTheColumnsAreHeadedByTheirThresholds: the columns are 95/90/80, and sizes 80, 89, 90, 94, 95, 100 of a limit of 100 land in the column their percentage heads;
• testEachColumnCountsTheKeysFromItsHeadingUpToThePreviousOne: for limits 1, 7, 100, 4000 and 4001 and every size up to the limit, a key is counted exactly when it holds at least 80%, and the column counting it is headed by a percentage the key reaches while the key stays below the percentage heading the column before it.

4/4 green. Five mutants of BackendStat are each caught: the bucket moved to 85% (the other way out the issue offers), > instead of >=, the columns in ascending order, a limit of 0 treated as a limit, and every near key reported in the last column.

The loop printing the headings is not covered by a unit test (it needs a backend); it reads the same list the thresholds do.

Note for #1132

The reference partial _variablelist-backendstat-index-status.adoc exists only in #1132, and names the column 85% there (while already describing the 80–90% range). Whichever of the two PRs is merged second updates that partial to 80%.

[maximthomas]

praise: The heading and the bucket now come from one list, and the new arithmetic changes nothing it shouldn't.

• NEAR_LIMIT_PERCENTS feeds both the headings (BackendStat.java:1035-1038) and nearLimitColumn, so the #1135 mismatch can no longer come from two separate literals.
• nearLimitColumn's size * 100 >= entryLimit * percent classifies exactly like the old entryLimit * 0.95 / 0.9 / 0.8 doubles: 0 differences over 376,500,012 boundary sizes (every limit 1..5,000,000 plus 20M random limits up to Integer.MAX_VALUE).
• testEachColumnCountsTheKeysFromItsHeadingUpToThePreviousOne walks every size up to limits 1, 7, 100, 4000 and 4001.