pub struct RuntimeReport {Show 14 fields
pub retained_accounts: usize,
pub managed_accounts: usize,
pub lingering_accounts: usize,
pub retiring_accounts: usize,
pub restarting_accounts: usize,
pub manager_restarts: u64,
pub unrecovered_grants: u64,
pub uncertain_acquires: u64,
pub counter_overflow: bool,
pub refill: Option<LeaseStats>,
pub snapshots: SnapshotStats,
pub accounting: WriterHealth,
pub sharding: ShardOccupancy,
pub contention: ContentionReport,
}Expand description
Instance-wide aggregate of the runtime’s accounts, refill, snapshot and
accounting counters, from RuntimeHandle::report. Safe to expose as
metrics: it carries no per-account labels except the bounded
ContentionReport::hottest list.
Fields§
§retained_accounts: usizeAccounts with a lease slot on this instance. Slots are kept for the process’s life because they carry irreversible overage spend, so this only grows.
managed_accounts: usizeAccounts with a live lease manager: running or lingering.
lingering_accounts: usizeOf managed_accounts, those no longer eligible and waiting out
InstanceRuntimeConfig::idle_account_linger.
retiring_accounts: usizeAccounts whose manager is releasing its leases and not yet joined.
restarting_accounts: usizeAccounts waiting out InstanceRuntimeConfig::manager_restart_backoff
after their manager exited unexpectedly.
manager_restarts: u64Manager restarts after an unexpected exit, summed across accounts. Nonzero means a lease manager died; a rising count means one keeps dying.
unrecovered_grants: u64Summed across accounts: see AccountReport::unrecovered_grants.
Nonzero is crash exposure — granted units that return only at TTL
reclaim (INVARIANTS.md 9).
uncertain_acquires: u64Summed across accounts: see AccountReport::uncertain_acquires.
counter_overflow: boolSome counter in this report, or one it aggregates, exceeded u64 and
is saturated. Totals are then lower bounds, never wrapped values.
refill: Option<LeaseStats>Refill counters summed across every account, including managers that
have since retired or died. None when the sum overflowed.
snapshots: SnapshotStatsThe snapshot task’s counters.
accounting: WriterHealthThe usage writer’s accounting health.
sharding: ShardOccupancyWhat this instance’s shard layout is carrying (GL-124).
Sharding buys one thing — a request-serving thread writing to lines no
peer writes — and that holds only while the affinities handed out do not
outnumber the shards. When it stops holding, the instance degrades
toward the unsharded cost while looking exactly like the contention the
layout was enabled to remove, so it is reported rather than left to be
inferred from latency. ShardOccupancy::is_crowded is the question;
docs/LOCAL_SHARDING.md is what to do about the answer.
contention: ContentionReportAdmission exchanges that lost a race to another core, per account: lease and overage debits and concurrency-gauge acquisitions (GL-134, GL-139).
Which accounts, if any, are hot enough on this instance that their
funding line is written from several cores at once — the condition
opt-in lease sharding exists for (docs/LOCAL_SHARDING.md). Cumulative
since the process started, so read it as a rate between reports.