Depth was observed at insert entry but its sample was only counted at
insert exit, so an insert still in flight contributed nothing to the
mean while completed inserts -- disproportionately the fast, shallow
ones -- all did. The reported mean understated queueing exactly when
queueing was worst: with every writer inside its first insert the gauge
was omitted entirely, while writers-in-flight correctly showed them all.
Depth and its sample count are now both folded in at entry, so the mean
is taken over one population. Mean depth is the L in Little's Law, so a
biased L understated the derived queueing share of each insert.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Drops nodestore_read_us and everything added to reach it. read_mean_us already
carries microsecond precision and separated the two sync failure modes cleanly
in live testing -- 8.8 us on a clean store against a 223 us cold-store peak --
so the distribution added no signal that changed a diagnosis.
The cost of getting it was disproportionate. NodeStoreScheduler had no path to
the metrics registry, so its production constructor grew a ServiceRegistry
parameter: a metric addition changing a production signature. That in turn
forced an edit to a pre-existing test, src/test/app/SHAMapStore_test.cpp, whose
only stake in this is that it constructs a scheduler. Worse, the scheduler is
built in Application's member initializer list, long before metricsRegistry_
exists, so the registry could not be captured once and had to be re-resolved on
every fetch -- a lookup on a path that runs millions of times per sync.
The constructor returns to taking JobQueue& alone and SHAMapStore_test.cpp
returns to the single-argument call, leaving that file differing from its
pre-change form only by the NodeStore:: to node_store:: rename it picked up from
develop.
FetchReport::elapsed stays microseconds and onFetch keeps its explicit
duration_cast to milliseconds for addLoadEvents, which takes milliseconds. That
widening was a separate fix and is what makes read latency measurable at all.
kSubMillisecondBoundaries loses its only consumer and regains [[maybe_unused]],
which is the state the commit that introduced it left it in; without the
attribute an unused constant is an error under wextra with werr.
Also removes the ledger-data-sync panel that charted the histogram and the
fetch_type and found template variables, which filtered on labels no metric
emits any more, plus the runbook and reference-doc sections and the two
instrument and view counts that named it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
kSubMillisecondBoundaries existed but nothing used it, so per-fetch read
latency never reached Grafana -- only the coarse read_mean_us gauge did,
which cannot separate "every read took 9us" from "most took 2 and a few
took 900".
Add a nodestore_read_us histogram, register its view against the sub-
millisecond ladder rather than kMicrosecondBoundaries (whose first edge
is 100us, above the entire range a warm read occupies), and record into
it from NodeStoreScheduler::onFetch using FetchReport::elapsed, which a
previous change widened to microseconds for exactly this purpose.
The name and its labels live in a new include/xrpl/telemetry header
because the view registration (xrpld.telemetry) and the record site
(xrpld.app) sit in different levelization modules; a copy-pasted literal
would let them drift and silently drop the bucket override. Same reason
and same placement as GetObjectMetricNames.h. No new levelization edge:
xrpld.app > xrpl.telemetry already exists.
NodeStoreScheduler had no registry access, so it now takes a
ServiceRegistry and resolves the registry per call. It is constructed in
Application's initializer list, long before metricsRegistry_ is assigned
in setup() and started in startTelemetry(), so capturing a pointer at
construction would capture nullptr forever; the metric macros null-check
the registry, the meter and the instrument, so early fetches are simply
not recorded.
Labels are fetch_type and found, both already carried on the report --
4 series, fixed at compile time. A slow async read delays prefetch while
a slow sync read blocks a caller, and a miss can cost a read of every
backend, so neither dimension can be collapsed.
Negative elapsed times are skipped: the SDK rejects them and logs a
warning on every call, which on a per-fetch path is a log flood. Zero is
still recorded, since a page-cache-served read genuinely rounds to it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
FetchReport::elapsed was milliseconds, so every nodestore read rounded to
zero: a warm store answers in single-digit microseconds and a cold one in
low hundreds, and both became 0 ms. That difference is the whole signal
separating a cold-read stall from a healthy node, and it was being
discarded at the type. Database::fetchNodeObject now measures once and
uses that one value for both the cumulative counter and the report, so
the two can never disagree. The job-queue call still takes milliseconds
and now casts explicitly.
BatchWriteReport::elapsed stays milliseconds and is documented as such:
a batch write covers many objects and reaches the disk, so it belongs in
that range.
Also adds a sub-millisecond histogram ladder, because the existing bucket
edges start at 100 microseconds and put the entire warm range in bucket
0. It is not wired to a view yet: no sub-millisecond instrument exists to
name, so the edges wait for the instrument that records read latency.
The new test captures what the nodestore reports and asserts the reported
total equals the internal microsecond accumulator exactly, plus that at
least one report is not a whole number of milliseconds -- which a
millisecond-typed field can never satisfy on any hardware.
A saturated ledgerData lane makes TimeoutCounter re-arm its timer
without running the timer body, so timeouts_ never advances and the
six-timeout give-up can never fire. Acquisitions then neither finish
nor fail until the one-minute sweep destroys their partial maps, and
the work restarts. Every step of that chain was debug-log-only, so a
node at warning level could not be diagnosed after the fact.
The counters are separate on purpose: deferrals rising while timeouts
stay flat is the signature, and no single counter shows it.
Completions are recorded in done() rather than at the "Done: complete"
log line, because that line also fires for failures and misses the
checkLocal and receiveNode paths; done() is the one funnel every
outcome passes through and its signaled_ guard makes it idempotent.
AcquireStats is only forward-declared in ServiceRegistry so libxrpl
still includes nothing from xrpld. The src/ include path for the test
binary moves out of the telemetry guard, since a header-only type
under src/xrpld/ is testable in every build.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The consensus headers moved into the isolated xrpl/consensus module and
took a dependency on xrpl/telemetry for the tracing span constants, but
two things were left behind:
- Four includes still pointed at the old src/xrpld/consensus/ location,
which no longer exists, so the build failed with
"fatal error: 'xrpld/consensus/ConsensusParms.h' file not found".
- xrpl.libxrpl.consensus never linked xrpl.libxrpl.telemetry. add_module
isolates each module's headers, so xrpl/telemetry/SpanNames.h was not
on the include path even once the include was repointed.
Repoint the stale includes at xrpl/consensus/, and declare the telemetry
module before consensus so consensus can link it. Regenerate ordering.txt
for the resulting edge.
NuDB serializes every insert behind one global mutex held for the whole
call, so a caller cannot see how long it waited. Record instead the
writer depth joined at and the wall time spent; with mean depth L and
mean insert time W, Little's Law gives service time W/L and queueing
W - W/L. That distinguishes a serialized write path from a saturated
disk: measured on a dev box the device sat 89 percent idle while
throughput stayed flat at 42k inserts per second.
The accounting runs from a ScopeExit guard because the insert can
allocate and therefore throw; leaking the depth would strand the gauge
above zero for the life of the process.
getWriteLoad also stops returning a hardcoded zero. It now reports
writer depth, which is bounded by the writing-thread count and so stays
far below the kMaxWriteLoadAcquire cutoff that gates history
acquisition, where returning bytes or microseconds would have silently
suppressed it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
fetchDurationUs_ had no getter, so telemetry reached it by building a JSON
object, stringifying a uint64 and parsing it back with stoll on every
collect tick. storeDurationUs_ was declared and never written or read at
all, along with the jss::node_writes_duration_us key.
Both now have accessors, both production store paths time their backend
call, and the registry reads them without the round trip. get_counts also
reports the write duration, so the RPC and the metric agree. Mean read
latency is the signal that separates a cold store from a warm one: warm
reads are single-digit microseconds, cold ones low hundreds.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Both were uint32_t while every sibling counter was uint64_t. On a
multi-day node node_read_bytes read 1,481,244,491 for 1,894,924,394
reads, i.e. 0.8 bytes per read, which is impossible: the counter had
wrapped about 350 times. Read hit rate is used to tell a cold-read
stall from a write-lock ceiling, so a wrapping numerator makes that
diagnosis wrong rather than merely imprecise.
getFetchTotalCount() was already backed by a uint64_t member, so its
32-bit return type truncated a correct value at the accessor. All
three getters now return uint64_t.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Slowness on the peer object-fetch path could be observed but not
attributed. Job duration metrics carry only `job_type`, and both
`RcvGetLedger` and `RcvGetObjByHash` report as `ledgerRequest`, so a
queue-wait spike could not be traced to a handler. Nothing measured
NodeStore cost, request size, or the differential charge.
Latency now decomposes into three additive parts, each separately
measurable:
end-to-end = queue wait + NodeStore lookup + everything else
- `handler` label on job_queued_total/_started_total/_finished_total and
job_queued_us/job_running_us. The value is sanitised: a name passes
through only if non-empty and all ASCII letters, else "other". Two job
names embed a ledger sequence, so a raw label would mint one series
per ledger; the rule bounds the domain at 43 names plus "other".
- getobject_lookup_us, _request_objects, _lookups_total{result},
_rejected_total{reason} and _charge, recorded at their call sites.
All three histograms get explicit bucket views: the SDK default stops
at 10,000, which every one of them exceeds.
- Per-job-type waiting/running/deferred gauges for the 35 non-special
job types. `deferred` is the leading indicator, since addJob never
rejects -- it defers, so backpressure otherwise shows up only as
latency after the fact.
`JobQueue::collect()` snapshots the counters under the queue lock and
publishes gauges after releasing it. Writing them while holding the lock
would invert a lock order against the collector's own lock, which the
collector's flush thread already holds when it calls this hook.
Tests assert exact values, including that the charge is priced on the
requested count rather than the capped iteration count.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Add shared current_ledger_seq / current_ledger_hash span attributes so a
transaction's work can be joined to the ledger trace that produced it, and
fix discrepancy D1 (txq.enqueue was a detached trace root).
- Define current_ledger_seq / current_ledger_hash once in SpanNames.h and
re-export via `using` from TxQ/TxApply/Tx span-name headers. These name the
ledger being worked on (open/tentative apply or in-flight consensus build),
distinct from ledger_seq (the built/validated ledger on ledger.build /
consensus.round). Named after the RPC field ledger_current_index.
- txq.enqueue: set current_ledger_seq/hash from the view, and parent the span
to the caller's tx.process span via an explicit captured SpanContext (new
trailing TxQ::apply param) instead of a detached root. The parent is
explicit, not ambient-inherited, and the ScopedSpanGuard scope is RAII-bound
to the synchronous apply, so it cannot leak onto a reused worker (D1 fix).
On the open-ledger rebuild path no tx.process context exists, so it stays a
root and the attribute provides the correlation.
- tx.preclaim / tx.transactor: set both attributes from their ledger view.
tx.preflight is stateless (no view) and is the documented exception.
- tx.process / tx.receive: set current_ledger_seq from the current open ledger
index at submit/receive time (no hash: not yet applied to a ledger).
- Contract test pins the two new attribute key strings.
Neither key is a spanmetrics dimension, so there is no metric-cardinality
impact. Dashboards/collector/docs land on the later phases per the chain split.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
