QuestDB has no multi-table transaction, so a HistoryUpdate writes its trail
row before the sample change. A failed or crashed change left an orphan
trail row that modified reads served as a change that happened, and a trail
INSERT re-sent after a client timeout duplicated (the table had no key).
- sample_modified gains `pending` and DEDUP UPSERT KEYS(ts, tag,
modified_id). The trail row is written pending; after the change lands
the same row is re-upserted pending = false (the applied marker).
- A transient trail INSERT failure is re-sent once with the same ids; a
first attempt that did apply converges on the key.
- ReadModifiedAsync serves a pending row only when it is the latest trail
row at its time and the sample there was written at or after the
modification (a crash between change and marker). The orphan of a change
that never landed, typically followed by the client's retry, is dropped.
ASOF JOIN over the window's samples of one tag; the resume filter sits
outside the window function so it never cuts a time's partition.
- The bootstrapper migrates an older trail (ADD COLUMN pending, DEDUP
ENABLE) and waits for it to apply (bounded 10 s); old rows read
pending = false and stay visible. Until applied, or with
AutoCreateSchema=false, the old format is used.
Rebased from the unpushed remed/lane-H2 branch (originally authored 2026-10-22
in session_01Ub9hGmYLQvn9XETzw7CMsT) onto current master.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
Event reads filtered on source and every modified-value read filtered on tag
scanned that symbol column in every partition of the window; only
sample.tag carried an index (#683).
Both columns are now SYMBOL INDEX in the DDL, and the bootstrapper's index
migration is generalised over QuestDbTables.SymbolIndexes (sample.tag,
event.source, sample_modified.tag): a missing index is added with the same
idempotent ALTER ... ADD INDEX, Warning-only when AutoCreateSchema is false
or the ALTER is refused. Integration tests: new tables carry all three; a
pre-index table set gains them on boot and event reads answer the same.
This reverses #683's documented choice to leave the two unindexed;
docs/Historian.md carries the reasoning.
Rebased from the unpushed remed/lane-H2 branch (originally authored 2026-10-22
in session_01Ub9hGmYLQvn9XETzw7CMsT) onto current master.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
The QuestDB live writer waited for WAL visibility after every tag group, and
the wait polled wal_tables() every 10 ms. With DrainConcurrency = 8 that was
up to eight pollers at ~100 queries/s each against a database whose WAL
apply was already behind.
New Runtime seam IHistorianPassWriter (WriteLiveValuesInPassAsync +
FlushAsync). The recorder writes every group of a pass through it and calls
FlushAsync once after all groups report, before the ack, whenever any group
wrote (an un-acked pass is redelivered and its insert-if-absent check must
see this pass's rows). A flush failure does not fail the pass. The QuestDB
writer implements it: in-pass writes only mark a flush as owed; FlushAsync
waits once. The poll now backs off 10 -> 20 -> ... -> 200 ms, never past the
timeout. Direct WriteLiveValuesAsync callers still wait per call.
Also updates the QuestDB cancelled-write integration test to the rethrow
contract from the previous commit.
Rebased from the unpushed remed/lane-H2 branch (originally authored 2026-10-22
in session_01Ub9hGmYLQvn9XETzw7CMsT) onto current master.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
Every IHistorianValueWriter caught Exception and returned false, including
OperationCanceledException from the drain's own token. The recorder's guard
("a cancelled pass is not a tag failure") was therefore dead code: at shutdown
mid-pass each cancelled group could be booked as a consecutive failure and,
with a sibling that already wrote, tip a healthy tag toward quarantine.
SqlServer, QuestDb and Gateway writers now rethrow a cancellation of the
caller's token; a cancellation-shaped error on a live token is still an
ordinary failed write.
The Gateway writer relabelled a Local-kind timestamp as UTC instead of
converting it, so the same value could land at two different times on two
backends. Stamped samples now go through LiveWritePreparer (Local -> UTC,
non-finite doubles not sent, duplicate timestamps keep the first); unstamped
samples are still left for the gateway to stamp, in batch order.
Rebased from the unpushed remed/lane-H2 branch (originally authored 2026-10-22
in session_01Ub9hGmYLQvn9XETzw7CMsT) onto current master.
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
QuestDB speaks plain PG-wire; Npgsql's default GSS Encryption Mode=Prefer made
Linux central boots log 'Cannot load library libgssapi_krb5.so.2'. An explicit
'GSS Encryption Mode' in the connection string is kept. Not re-verified on a
Linux host here; the Postgres Sql driver is left alone (Kerberos may be wanted).
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
Review R08:1.1: an uncapped HistoryRead-Raw (NumValuesPerNode = 0) read a
tag's whole window into memory on every backend (SQL TOP(int.MaxValue),
QuestDB LIMIT, Gateway max_values = int.MaxValue + a second snapshot copy).
New ServerHistorian:MaxRawValuesPerRead (default 10 000; <= 0 unbounded with
a Validate() warning; 1 raised to 2). Each backend data source takes it,
reports it as IHistorianDataSource.MaxRawValuesPerRead (the Gateway: min
with a strict gateway's learned ceiling), so the node manager's existing
RawPageLimit paging serves N = 0 and N > cap as pages of at most the cap
with a continuation point. Independently, a backend raw read asking for 0
or more than the cap reads cap + 1 rows and is REFUSED
(HistoryReadLimitExceededException => BadTooManyOperations) when the window
holds more - never answered truncated, since the server's paging reads a
short answer as the last page. Bound through the Host registry, the
SqlServer/QuestDb providers and GatewayHistorian.CreateDataSource.
Tests: options (default/binding/warnings), provider binding, Gateway unit +
through a real SDK node manager (uncapped -> 999/998/998/605 + CPs), and a
conformance case for the SQL backends.
Refs #736
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
QuestDB: {p}sample.tag is now SYMBOL INDEX. The processed-read data-range
lookups (EarliestSample / LatestSample) were partition scans of the unindexed
symbol; on 53.6M rows / 120 tags / 61 daily partitions the indexed seeks are
1-1.5 ms warm / 2-4 ms cold whatever the tag's history (unindexed: up to 11 ms
warm, 14 ms cold; cost grew with the table). The bootstrapper adds the index to a
pre-#683 table (ALTER ... ADD INDEX, a WAL txn, idempotent, ~3 s for 68M rows);
AutoCreateSchema=false only logs the statement. Disk +~12%; write apply time
within noise; main per-sample/raw/time-weighted queries unchanged or faster.
Rejected by measurement/analysis: bounded existence probes (still full scans
when a side is absent), LATEST ON (slowest unindexed), per-tag first/last side
table (cannot track TTL drops or deletes INSERT-only).
Both backends now read the data range before the bucket statement: an append
racing the read reaches the buckets but not the range, so it can no longer
produce a Good Count of 0 for an interval that has samples.
Tests: TagIndexTests (new-table index, bound-param seek uses the index,
tombstones excluded / outage marker counted, pre-#683 migration with identical
processed results across 10 aggregates x 4 windows x both directions).
- Average/Minimum/Maximum/Range/*ActualTime refuse (BadAggregateInvalidInputs) only a GOOD
String/DateTime sample (new GoodNonNumericCount / good_non_numeric_count column); an Uncertain
one is skipped and makes the min/max family Uncertain, matching the SDK (Average skips it,
ComputeMinMax flags the uncastable value). Delta keeps its usable-sample refusal.
- ProcessedConformance: an Uncertain value above the Good maximum (pins UncMax/unc_max through
the real SQL) and an Uncertain String through the real SQL.
- The_uncertain_dataset_is_not_vacuous also proves the min/max STATUS discrimination.
Co-Authored-By: Claude <noreply@anthropic.com>
Part 13 defines Average/Minimum/Maximum/Range/*ActualTime over Good raw values
(independent of TreatUncertainAsBad) and the SDK's calculators skip Uncertain values
under the server's TreatUncertainAsBad=false. SqlServer/QuestDB value columns now use
the good flag (g) instead of usable (u); new UncMin/UncMax columns feed the Part 13
Minimum/Maximum/Range status rule (Uncertain on a Bad sample or an Uncertain value
beyond the Good extreme -- the SDK applies only the Bad half). An Uncertain-only
interval is BadNoData. Uncertain-bearing parity datasets (in-memory + real SQL) match
the SDK unmasked; the beyond-extreme divergence is pinned explicitly.
Co-Authored-By: Claude <noreply@anthropic.com>
QuestDB Start/End/Delta reads of a tag being written intermittently failed
BadHistoryOperationUnsupported (InvalidCastException: Column 'ts_3' is null):
the StartEnd query referenced its CTE four times and QuestDB opens a table
reader per reference, so a row committed between two of them was counted but
not picked. MinimumActualTime/MaximumActualTime had the same race and returned
a value-less Good.
(a) SQL
- QuestDB StartEnd: one scan, one GROUP BY — first()/last() for the
any-quality picks, arg_min/arg_max keyed on the usable rows' time for the
usable picks (no str_value: arg_min has no VARCHAR form; only Delta reads
them). Byte-identical output to the old query on a 1M-row tag, ~15% faster.
- SqlServer StartEnd: one windowed pass + conditional MAX (was five CTE
references); identical output, 1.4-3x faster.
- ActualTime kept on both backends: every single-scan form measured 1.2-2.4x
slower; its inconsistency is absorbed by (b).
(b) Defence in depth on both backends
- ReadBucket takes DbDataReader and reads absent picks as null
(ReadStoredOrNull) instead of throwing.
- AggregateEvaluator: Delta with a missing first/last usable pick and
ActualTime with a missing extreme are BadNoData (were BadAggregateInvalidInputs
/ a value-less Good); Start/End with no pick already were.
- Unit tests (Sql.Common evaluator, QuestDb + SqlServer mappers over a
DataTableReader) fail with the guards removed.
LiveWriteReadRaceStressTests (QuestDB integration, opt-in via
HISTORIAN_QUESTDB_STRESS_SECONDS, Category=SlowUnit): single-row writer vs
4 tight-loop readers. Old code: 6 faults in ~105k reads; new: 0 in 106k.
Docs: Historian.md QuestDB processed-read note; F2 (Min/Max on the interval
start: SDK raw, ours Calculated) and F3 (integer tags: SDK casts interpolated
values to the integer type, ours Double) added to the SDK-parity deviations.
Every HistoryUpdate a SQL backend applies now leaves a modification trail
(prior value, principal, Part 11 update type, modification time), and
HistoryReadRaw with IsReadModified=true serves it as HistoryModifiedData.
- Core: HistoryUpdateContext threaded into all three IHistorianDataUpdater
methods; IHistorianDataSource.ReadModifiedAsync (default NotSupported =>
BadHistoryOperationUnsupported on Gateway/Null); HistoryModified* contracts.
- SqlServer: schema v2 (002_sample_modified.sql, RequiredVersion=2, docs copy);
trail written by OUTPUT ... INTO on the same MERGE/DELETE statement, so a
failed update leaves no audit row; retention purges SampleModified per tag on
the source-time cutoff; ReadModifiedAsync keyed on (SourceTimeUtc, ModifiedId).
- QuestDB: append-only {prefix}sample_modified (TTL like samples), trail row
written BEFORE the change (orphan-on-crash, never an untracked change);
existence reads now carry the prior value; ReadModifiedAsync.
- Node manager: modified reads dispatch + page via the #669 continuation store
(HistoryContinuationState gains HistoryCursorKind + LastModifiedId), gated by
the existing HistoryRead decision; session principal / key id passed to the
updater. IsDeleteModified stays unsupported.
- Client.CLI: historyread --modified.
- Tests: node-manager dispatch/paging/cursor binding/principal threading,
over-the-wire HistoryModifiedData, shared ModifiedValueConformance on both
backends, SqlServer atomicity (forced trail failure => no change) + v1->v2
upgrade + retention trail purge.
- Docs: Historian.md (Modified-value history section, schema v2 upgrade note),
security.md, Client.CLI.md, CLAUDE.md.
- I-1: QuestDbConnection caps Npgsql Maximum Pool Size at 24 (DefaultMaxPoolSize)
when the connection string does not set it; an explicit value is kept. The
effective value is logged once at Information.
- M-1: QuestDbTagCatalog.EnsureAsync waits for the tag table's sequenced commits
and re-checks before inserting, so it no longer overwrites a provisioner row
(unit/description) that is committed but not yet applied. Integration test
reproduces the clobber against real QuestDB (falsified).
- M-3: the pre-1970 drop Warning is rate-limited per tag (once per 10 min);
DroppedPre1970 still counts every sample.
- M-5: QuestDbHistorianProvider is IAsyncDisposable/IDisposable and disposes the
shared Npgsql pool; Program.cs registers the provider through a factory and
every seam resolves it from the container, so DI disposes it at shutdown.
- M-8, M-9: docs fixes (HistoryUpdate transaction wording; folded duplicate
HistoryUpdate paragraph). Pool-size docs updated for the new default cap.
Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe
Joseph Doherty