The cron-owned SLA sweep, verbatim-shared by the hosts: sweepSLA over the
SYSTEM store (the only legitimate cross-tenant read+write, unreachable
over HTTP), audit to logs + optional queue, a structured line per
escalation, and containment — a sweep failure logs a maintenance-error and
resolves rather than throwing, so a caller running other duties is never
starved by this one (the two-cron split covers the uncatchable CPU-limit
case). Queue sends are collected and awaited HERE: the whole runner
already executes under the handler's ctx.waitUntil, where a nested
waitUntil is unavailable.
The cron-owned SLA sweep, verbatim-shared by the hosts: sweepSLA over the SYSTEM store (the only legitimate cross-tenant read+write, unreachable over HTTP), audit to logs + optional queue, a structured line per escalation, and containment — a sweep failure logs a maintenance-error and resolves rather than throwing, so a caller running other duties is never starved by this one (the two-cron split covers the uncatchable CPU-limit case). Queue sends are collected and awaited HERE: the whole runner already executes under the handler's ctx.waitUntil, where a nested waitUntil is unavailable.