fix(boost): fix seven correctness bugs in boost poller

[somethingwithproof]

Summary

Fixes the BOOST ERROR: Failed to retrieve archive table name that appeared at exactly the same time on consecutive days (replication lag race condition), plus ten additional correctness bugs and five defensive programming improvements found during code review.

Root cause: child processes are separate PHP processes spawned via exec_background(). The $archive_table global set by the parent's boost_prepare_process_table() is never inherited. Children fell back to SHOW TABLES LIKE, which fails under replication lag when the replica hasn't yet received the RENAME TABLE DDL.

Correctness fixes:

• --archive-table CLI argument: parent passes the archive table name to each child; child validates with /^poller_output_boost_arch_\d+$/ before accepting
• boost_get_arch_table_names: removed AND TABLE_ROWS > 0 from the information_schema fallback; InnoDB's TABLE_ROWS is an estimate and can be 0 immediately after RENAME TABLE
• sig_handler: gate boost_poller_status = 'terminated' write on !$child; children were clobbering the parent's status entry
• boost_prepare_process_table: guard intval($processes) <= 0 before ceil() division; boost_parallel unset causes division by zero
• boost_output_rrd_data: return 0 not false for both early-exit paths (no arch tables, zero total rows) to keep the status column numeric; guard $max_per_select <= 0 with fallback of 1000
• startup barrier: wait up to 30 s for the first child to register before the monitoring loop; exec_background() is non-blocking and the loop exited prematurely when children hadn't yet called register_process_start()
• boost_process_poller_output: remove static $archive_table = false that shadowed the global populated by the new CLI argument; add $archive_table to the global declaration
• non-templated DS queries in boost_process_local_data_ids: add LEFT JOIN graph_templates_item AS gti ON dtr.id = gti.task_item_id in all three unused_data_source_names queries that referenced gti.task_item_id without joining the table (MySQL raised "Unknown column")
• boost_poller_status on $continue = false: boost_prepare_process_table() sets status to 'running' before returning false; add an else branch to reset it to 'complete' so the next run does not emit a false overrun warning
• boost_get_total_rows: replace SUM(TABLE_ROWS) InnoDB estimate with exact COUNT(*) per table; TABLE_ROWS can report 0 immediately after a poller flush, causing boost to skip the RRD update run entirely
• orphaned arch tables: drop zero-row arch tables before returning false so subsequent runs don't pick up empty tables

Defensive programming:

• Escape device-sourced output in bulk INSERT: apply db_qstr() to all string fields in boost_poller_on_demand; device data retrieved from the DB was concatenated raw into the INSERT statement
• Log path character allowlist: validate path_boost_log against [A-Za-z0-9_./-] before placing it in a shell redirect argument; exec_background redirect args bypass escapeshellarg by design
• Array args to exec_background: pass child process arguments as an array so each element receives cacti_escapeshellarg() individually rather than as a concatenated string
• Backtick-quoted dynamic table names: wrap all dynamically-generated table names in backticks across both files to guard against reserved-word collisions
• Cache file permissions: replace chmod($cache_file, 0666) with 0644; replace mt_rand() with random_int() for temporary table name suffix

Test plan

• php -d error_reporting=24575 include/vendor/bin/pest tests/Unit/BoostProcessBugsTest.php tests/Unit/BoostArchiveTableSmokeTest.php — 30 tests, 80 assertions, all pass
• BoostArchiveTableSmokeTest (8 tests): PHP syntax, table name regex, injection rejection, exec_background argument presence
• BoostProcessBugsTest (22 tests): all original regression tests plus 12 new tests for the fixes in this PR
• Manual: enable boost on a replicated Cacti instance and verify no BOOST ERROR: Failed to retrieve archive table name in cacti.log

[Copilot]

Pull request overview

This PR bundles seven correctness fixes for the Cacti boost poller, motivated by a "Failed to retrieve archive table name" error caused by a race between RENAME TABLE and replica lag. Child processes are now told the archive table name on their command line and validate it before use, plus several adjacent reliability issues (status clobbering by children, divide-by-zero on misconfigured boost_parallel, premature exit of the monitoring loop, a static-shadows-global bug in boost_process_poller_output, and a broken JOIN in the non-templated DS path) are addressed. Two new Pest test files assert these invariants via source-text inspection.

Changes:

• Pass and validate --archive-table= to child workers, declare $archive_table global in boost_launch_children/boost_process_poller_output, and drop the TABLE_ROWS > 0 filter / static shadowing in lib/boost.php.
• Harden poller_boost.php: parent-only boost_poller_status='terminated' write, intval+guard on boost_parallel, boost_output_rrd_data returns 0 (not false) and guards max_per_select, add a 30s startup barrier after boost_launch_children(), drop orphaned arch tables when total rows is 0, replace info-schema TABLE_ROWS with SELECT COUNT(*) in boost_prepare_process_table, fix non-templated unused_data_source_names queries with a LEFT JOIN graph_templates_item.
• Add tests/Unit/BoostProcessBugsTest.php and tests/Unit/BoostArchiveTableSmokeTest.php exercising the above invariants.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

File	Description
poller_boost.php	Adds --archive-table CLI arg, sig_handler parent-only guard, intval/zero-guard on boost_parallel, startup barrier, return 0 on empty rows, max_per_select guard, orphan-table cleanup, help text.
lib/boost.php	Removes TABLE_ROWS > 0 filter, removes static shadowing of $archive_table, adds LEFT JOIN graph_templates_item to non-templated unused_data_source_names queries.
tests/Unit/BoostProcessBugsTest.php	New regression tests verifying each source-level invariant of the fixes.
tests/Unit/BoostArchiveTableSmokeTest.php	New smoke tests for PHP parse validity, archive-table regex, and exec_background argument plumbing.

Comments suppressed due to low confidence (1)

poller_boost.php:539

• boost_launch_children still reads boost_parallel without the intval() cast / <= 0 guard that was just added to boost_prepare_process_table (line 496-500). The two functions now use inconsistent semantics for the same config option: a value like "0" or a non-numeric string would be coerced to 1 here (via empty()), but in boost_prepare_process_table it is coerced to intval(...) <= 0 ? 1 : intval(...). For consistency and to keep for ($i = 1; $i <= $processes; $i++) safe against weird config values, apply the same intval() + <=0 pattern here.

	$processes = read_config_option('boost_parallel');

	if (empty($processes)) {
		$processes = 1;
	}

[TheWitness]

There can be multiple archive tables, not just one. It's an old legacy issue. However, there are numerous style changes in addition to the change itself that make peeling away the real change more difficult.