Skip to content

Use pg background worker, get rid of CREATE EXTENSION for oracle #74

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
serprex wants to merge 2 commits into from
Open

Use pg background worker, get rid of CREATE EXTENSION for oracle #74

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
fb4ad96
Use pg background worker, get rid of CREATE EXTENSION for oracle
serprex Jul 28, 2026
d9b928a
understand ddl inside transaction, allowing more mixed DDL/DML transa…
serprex Jul 28, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
36 changes: 4 additions & 32 deletions .github/workflows/ci.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -138,9 +138,10 @@ jobs:
fixtures/wal/xlog_switch/capture.sh
fixtures/wal/vacuum_full_pg_depend/capture.sh

# Install the walshadow extension before the e2e suite so each shadow PG
# can CREATE EXTENSION walshadow for the oracle decode path.
- name: Build + install walshadow extension (e2e oracle path)
# Build walshadow.so before the e2e suite: the bridge tests point
# dynamic_library_path at the build tree, so `make` alone is what they
# need. `make install` additionally proves the target docker relies on.
- name: Build + install walshadow PG module (e2e oracle path)
working-directory: walshadow/pgext
run: |
PG_CONFIG=/usr/lib/postgresql/${{ matrix.pg-major }}/bin/pg_config
Expand Down Expand Up @@ -184,35 +185,6 @@ jobs:
if-no-files-found: ignore
retention-days: 7

# walshadow extension is a separate PGXS module (shipped to shadow PG).
# Build it against the matrix PG, install into that major's $libdir /
# share/extension, then run pg_regress under a private --temp-instance
# so we don't need an external server. Covers the C bridge function's
# varlena, fixed-width by-val, by-ref, NULL, and ereport paths.
- name: walshadow extension (build + pg_regress)
working-directory: walshadow/pgext
run: |
PG_BIN=/usr/lib/postgresql/${{ matrix.pg-major }}/bin
PG_CONFIG=$PG_BIN/pg_config
make PG_CONFIG=$PG_CONFIG
sudo make PG_CONFIG=$PG_CONFIG install
PG_REGRESS=/usr/lib/postgresql/${{ matrix.pg-major }}/lib/pgxs/src/test/regress/pg_regress
rm -rf /tmp/walshadow_pgext_tmpinst
"$PG_REGRESS" \
--inputdir=./ --outputdir=./ \
--bindir="$PG_BIN" \
--temp-instance=/tmp/walshadow_pgext_tmpinst \
--port=54329 \
walshadow

- name: walshadow extension regression diffs on failure
if: failure()
working-directory: walshadow/pgext
run: |
test -f regression.diffs && cat regression.diffs || true
test -d /tmp/walshadow_pgext_tmpinst/log && \
sudo tail -n 200 /tmp/walshadow_pgext_tmpinst/log/postmaster.log || true

- name: ClickHouse server logs on failure
if: failure()
run: sudo tail -n 500 /var/log/clickhouse-server/clickhouse-server.err.log || true
Expand Down
20 changes: 13 additions & 7 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -66,16 +66,19 @@ Binaries land under `target/release/`:
- `walshadow-filter`, segment-level filter for offline WAL files
- `walshadow-classify`, record-level classifier for diagnostics

The PG extension under `pgext/` is built separately via PGXS, only
needed for the decode oracle:
The PG module under `pgext/` is built separately via PGXS. It backs the
decode oracle and the catalog overlay reads:

```
make -C pgext install
```

Loaded on shadow PG with `CREATE EXTENSION walshadow`. Absent extension
surfaces as `oracle fallback=N` on the status line; the daemon ships
raw on-disk bytes for `PgPending` types without it
Not an SQL extension: shadow's catalog is a read-only physical copy of
source's, so there is nowhere to run `CREATE EXTENSION`. Module must be
installed in shadow PG and loaded through `shared_preload_libraries`. Daemon
writes preload and `walshadow.*` settings for shadows it owns, then requires
worker socket. `--bridge-socket` defaults to
`<shadow-socket-dir>/walshadow-bridge.sock`

## Running standalone

Expand Down Expand Up @@ -128,12 +131,15 @@ params stay boot-only
## Testing

```
make -C pgext
cargo test
cargo clippy --all-targets -- -D warnings
```

Integration tests under `tests/` need `initdb` + `pg_ctl` on `PATH`
and spin a transient shadow PG per case. Walshadow-side timeouts are
and spin a transient shadow PG per case. Every shadow preloads the
module out of `pgext/`, so build it first: tests fail rather than skip
without it. Walshadow-side timeouts are
seconds-scale by design — long timeouts mask stalls rather than
surface them

Expand All @@ -143,7 +149,7 @@ surface them
src/ walshadow daemon + library
src/bin/ CLI entry points (stream, filter, classify)
clickhouse-c-rs/ CH-Native client, separate submodule
pgext/ walshadow decode-bridge extension (PGXS)
pgext/ walshadow decode-bridge PG module (PGXS)
architecture/ overview + internals diagrams
plans/ component design docs (overview.md is the baseline)
docker/ docker-compose demo + Dockerfile
Expand Down
76 changes: 38 additions & 38 deletions architecture/oracle.dot
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,20 +1,19 @@
// walshadow — PgPending decode oracle
// PgPending resolver path fronted by libpq into shadow PG. Extension
// surface (pgext/) shown as a side branch with absence-fallback
// semantics.
// walshadow, PgPending decode oracle
// PgPending resolver path over a unix socket into shadow PG's preloaded
// worker. Per-item and transport fallback stay visible.
//
// regeneration spec:
// sources of truth: plans/oracle.md · pgext/ · src/ oracle-path module (grep src/ for oracle|pending)
// subsumes: plans/oracle.md § resolver flow + extension surface + absence semantics
// subsumes: plans/oracle.md § resolver flow + worker surface + failure semantics
// quality bar:
// - two-way branch on extension availability legible (not a tangle)
// - libpq round-trip reads as one canonical orange edge in/out of shadow
// - required bridge path stays legible
// - batched socket round-trip reads as one canonical orange edge
// - fallback path visually distinct from main path
// shared style: palette.md
digraph oracle {
rankdir=TB;
compound=true;
graph [fontname="Helvetica", labelloc="t", label="walshadow oracle — PgPending decode bridge", fontsize=14, splines=spline, nodesep=0.35, ranksep=0.9, bgcolor="#272623", fontcolor="#ECE1D7"];
graph [fontname="Helvetica", labelloc="t", label="walshadow oracle, PgPending decode bridge", fontsize=14, splines=spline, nodesep=0.35, ranksep=0.9, bgcolor="#272623", fontcolor="#ECE1D7"];
node [fontname="Helvetica", fontsize=10, shape=box, style="rounded,filled", color="#6E6963", fontcolor="#ECE1D7"];
edge [fontname="Helvetica", fontsize=9, color="#c1a78e", fontcolor="#ECE1D7"];

Expand All @@ -29,26 +28,26 @@ digraph oracle {

// ════════ ② oracle decode path ════════
subgraph cluster_oracle {
label="② oracle path — post-plan, decode workers call directly; metrics-only runs skip it"; style="rounded,filled"; color="#4c4641"; fillcolor="#34302c"; fontcolor="#ECE1D7";
resolve [label="resolve_pending_tuple\nwalk columns,\nPgPending | Unsupported\n→ resolve_pending\nswap to Text on Ok(Some)\nbest effort: shadow may lag\nrow's catalog interval", fillcolor="#5D3F40"];
stats [label="OracleStats\nresolved / fallback_raw /\nerrors", fillcolor="#5D3F40", shape=note];
label="② oracle path, post-plan; metrics-only runs skip it"; style="rounded,filled"; color="#4c4641"; fillcolor="#34302c"; fontcolor="#ECE1D7";
resolve [label="resolve_pending_tuple\ncollect PgPending | Unsupported\none DECODE batch per tuple\nswap each text answer to Text\nbest effort: shadow may lag\nrow's catalog interval", fillcolor="#5D3F40"];
stats [label="OracleStats\nresolved / fallback_raw / errors\n\nBridgeStats\nup / requests / errors /\nlatency / reconnects / items", fillcolor="#5D3F40", shape=note];
}

// ════════ ③ libpq client (own pool, not catalog's) ════════
client [label="Oracle::client\ntokio_postgres::Client\nSELECT walshadow_decode_disk(\n $1::oid, $2::bytea)\nreconnect-once on is_closed", fillcolor="#4D4D28"];
// ════════ ③ bridge client ════════
client [label="Bridge\ntokio::net::UnixStream\nHELLO version gate\nframed DECODE request\nreconnect + retry on\ntransport failure", fillcolor="#4D4D28"];

// ════════ ④ shadow PG + extension ════════
// ════════ ④ shadow PG + preloaded worker ════════
subgraph cluster_shadow {
label="④ shadow PG"; style="rounded,filled"; color="#4c4641"; fillcolor="#34302c"; fontcolor="#ECE1D7";
ext [label="walshadow extension\n(pgext/walshadow.c)\nSearchSysCache1(TYPEOID)\nDatum reconstruct ×4 branches:\n typlen=−1 varlena\n typlen=−2 cstring\n typbyval fixed (memcpy)\n fixed by-ref (palloc)\nOidOutputFunctionCall(typoutput)", fillcolor="#3D4128"];
probe [label="probe_extension\nSELECT EXISTS\n(pg_proc.proname =\n 'walshadow_decode_disk')\nat connect + reconnect", fillcolor="#3D4128", shape=note];
lc [label="pinned locale\nlc_numeric / lc_time = C\n(shadow bootstrap)", fillcolor="#3D4128", shape=note];
worker [label="walshadow background worker\n(pgext/worker.c)\nunix socket, mode 0600\none transaction per request\none subtransaction per item\nerror frame or connection close", fillcolor="#3D4128"];
decode [label="ws_decode_datum_text\n(pgext/decode.c)\nSearchSysCache1(TYPEOID)\nDatum reconstruct ×4 branches:\n typlen=−1 varlena\n typlen=−2 cstring\n typbyval fixed (memcpy)\n fixed by-ref (palloc)\nOidOutputFunctionCall(typoutput)", fillcolor="#3D4128"];
env [label="pinned output environment\nTimeZone = UTC\nDateStyle = ISO, MDY\nIntervalStyle = postgres\nextra_float_digits = 1\nbytea_output = hex", fillcolor="#3D4128", shape=note];
}

// ════════ ⑤ pgext build artifact ════════
subgraph cluster_pgext {
label="⑤ pgext/ (PGXS build)"; style="rounded,filled"; color="#4c4641"; fillcolor="#34302c"; fontcolor="#ECE1D7";
artifact [label="walshadow.so\nwalshadow.control\nwalshadow--0.1.sql\nCREATE FUNCTION\n STRICT IMMUTABLE", fillcolor="#4D3850", shape=note];
artifact [label="walshadow.so\ndecode.o + overlay.o + worker.o\nno control file\nno SQL script\nno pg_proc row", fillcolor="#4D3850", shape=note];
}

// ════════ ⑥ emit sinks (downstream branch off oracle) ════════
Expand All @@ -60,50 +59,51 @@ digraph oracle {
}

// ─── main flow edges (LR spine, high weight) ───
pend -> resolve [color="#A1A9CC", penwidth=2, label="always", weight=10];
unsup -> resolve [color="#A1A9CC", style=dashed, label="always", weight=5];
resolve -> client [color="#CBA85E", penwidth=2, label="resolve_pending\nper pending col", weight=10];
client -> ext [color="#CBA85E", penwidth=2, dir=both, arrowtail=open, lhead=cluster_shadow, label="libpq SELECT\ntokio_postgres pool\n(separate from\n ShadowCatalog)", weight=10];
pend -> resolve [color="#A1A9CC", penwidth=2, label="required bridge", weight=10];
unsup -> resolve [color="#A1A9CC", style=dashed, label="required bridge", weight=5];
resolve -> client [color="#CBA85E", penwidth=2, label="one DECODE batch\nper tuple", weight=10];
client -> worker [color="#CBA85E", penwidth=2, dir=both, arrowtail=open, lhead=cluster_shadow, label="framed unix\nsocket request", weight=10];

// ─── local tiers never reach the oracle ───
tier12 -> emit_text [color="#BF8C5F", style=dotted, constraint=false, label="local render"];
tier3 -> emit_text [color="#BF8C5F", style=dotted, constraint=false, label="local render"];

// ─── shadow internal: extension uses pinned locale via typoutput ───
ext -> lc [style=dashed, color="#6E6963", constraint=false];
client -> probe [style=dashed, color="#CBA85E", constraint=false, label="connect /\nreconnect:\nprobe pg_proc"];
// ─── shadow internal: worker calls decode under pinned output GUCs ───
worker -> decode [color="#6E6963", label="each item"];
decode -> env [style=dashed, color="#6E6963", constraint=false];

// ─── pgext install edge ───
artifact -> ext [style=dashed, color="#6E6963", lhead=cluster_shadow, ltail=cluster_pgext, label="install:\n$libdir +\nextension dir;\nCREATE EXTENSION\nwalshadow"];
// ─── pgext preload edge ───
artifact -> worker [style=dashed, color="#6E6963", lhead=cluster_shadow, ltail=cluster_pgext, label="$libdir or build tree\nshared_preload_libraries\n+ walshadow.* GUCs"];

// ─── stats fan-in (off-axis, constraint=false) ───
resolve -> stats [style=dashed, color="#CBA85E", constraint=false, label="resolved++ /\nfallback_raw++ /\nerrors++"];
client -> stats [style=dashed, color="#CBA85E", constraint=false, label="request /\ntransport counters"];

// ─── emit disposition (off-axis fan-out from oracle) ───
resolve -> emit_text [color="#BF8C5F", penwidth=2, constraint=false, label="Ok(Some)\npending\n→ Text"];
resolve -> emit_raw [color="#BF8C5F", style=dashed, constraint=false, label="Ok(None)\next absent /\nNULL / error\n(PgPending)"];
resolve -> emit_fail [color="#BF8C5F", style=dotted, constraint=false, label="Ok(None)\n(Unsupported)"];
resolve -> emit_text [color="#BF8C5F", penwidth=2, constraint=false, label="text item\n→ Text"];
resolve -> emit_raw [color="#BF8C5F", style=dashed, constraint=false, label="item or transport error\n(PgPending)"];
resolve -> emit_fail [color="#BF8C5F", style=dotted, constraint=false, label="item or transport error\n(Unsupported)"];

// ─── Legend ───
legend [shape=plaintext, label=<
<TABLE BORDER="0" CELLBORDER="1" CELLSPACING="0" CELLPADDING="4">
<TR><TD COLSPAN="2" BGCOLOR="#34302c"><B>node fill</B></TD></TR>
<TR><TD BGCOLOR="#4D4128" WIDTH="20"></TD><TD>decoder ColumnValue</TD></TR>
<TR><TD BGCOLOR="#5D3F40"></TD><TD>oracle (src/ops/oracle.rs)</TD></TR>
<TR><TD BGCOLOR="#4D4D28"></TD><TD>libpq client (own pool)</TD></TR>
<TR><TD BGCOLOR="#3D4128"></TD><TD>shadow PG / extension</TD></TR>
<TR><TD BGCOLOR="#4D4D28"></TD><TD>bridge client</TD></TR>
<TR><TD BGCOLOR="#3D4128"></TD><TD>shadow PG worker</TD></TR>
<TR><TD BGCOLOR="#4D3850"></TD><TD>pgext/ build artifact</TD></TR>
<TR><TD BGCOLOR="#5D4628"></TD><TD>emit sink</TD></TR>
<TR><TD COLSPAN="2" BGCOLOR="#34302c"><B>edge colour</B></TD></TR>
<TR><TD><FONT COLOR="#A1A9CC"><B>━━</B></FONT></TD><TD>pending → resolver (always)</TD></TR>
<TR><TD><FONT COLOR="#CBA85E"><B>━━</B></FONT></TD><TD>libpq SELECT (MAIN wire)</TD></TR>
<TR><TD><FONT COLOR="#A1A9CC"><B>━━</B></FONT></TD><TD>pending → resolver</TD></TR>
<TR><TD><FONT COLOR="#CBA85E"><B>━━</B></FONT></TD><TD>unix socket (MAIN wire)</TD></TR>
<TR><TD><FONT COLOR="#BF8C5F"><B>━━</B></FONT></TD><TD>emit-time disposition</TD></TR>
<TR><TD><FONT COLOR="#6E6963"><B>┄┄</B></FONT></TD><TD>pgext install (filesystem)</TD></TR>
<TR><TD><FONT COLOR="#6E6963"><B>┄┄</B></FONT></TD><TD>pgext preload (filesystem)</TD></TR>
<TR><TD COLSPAN="2" BGCOLOR="#34302c"><B>policy</B></TD></TR>
<TR><TD COLSPAN="2" ALIGN="LEFT">Best effort: resolution runs post-plan;<BR ALIGN="LEFT"/>shadow's catalog may lag the row's<BR ALIGN="LEFT"/>interval in DDL edge cases — accepted<BR ALIGN="LEFT"/>to mostly support unknown types.</TD></TR>
<TR><TD COLSPAN="2" BGCOLOR="#34302c"><B>extension-absence</B></TD></TR>
<TR><TD COLSPAN="2" ALIGN="LEFT">probe_extension at connect →<BR ALIGN="LEFT"/>has_extension=false →<BR ALIGN="LEFT"/>resolve_pending Ok(None) →<BR ALIGN="LEFT"/>stats.fallback_raw++ →<BR ALIGN="LEFT"/>emitter writes raw bytes.<BR ALIGN="LEFT"/>No failure. CREATE EXTENSION<BR ALIGN="LEFT"/>+ reconnect re-enables.</TD></TR>
<TR><TD COLSPAN="2" BGCOLOR="#34302c"><B>bridge absence</B></TD></TR>
<TR><TD COLSPAN="2" ALIGN="LEFT">Unset socket or boot connect failure<BR ALIGN="LEFT"/>means no Oracle. PgPending reaches<BR ALIGN="LEFT"/>emitter unchanged and writes raw<BR ALIGN="LEFT"/>bytes. Unsupported remains fatal.<BR ALIGN="LEFT"/>Runtime transport failure retries<BR ALIGN="LEFT"/>once, then uses same disposition.</TD></TR>
</TABLE>
>];
ext -> legend [style=invis];
worker -> legend [style=invis];
}
Loading
Loading