Repository navigation
fix(exec): derive the query memory budget from the process limit and let joins spill or re-side instead of failing at a fixed 512 MiB #1985
Description
Part of #1982.
Problem
Query memory is a fixed 512 MiB DataFusion pool (SessionResourceConfig::default, crates/graphforge-exec/src/session.rs) with spill disabled by default, and hash joins cannot spill at all. Two failures at SF1 on frozen c09577e2e:
- IC3, every binding:
Resources exhausted: Additional allocation failed for HashJoinInput[7] with top memory consumers … HashJoinInput[1] (can spill: false) consumed 58.4 MB …. - BI1 on perf(query): property-equality lookups and expansions cost O(dataset) per query on durable projects #1931's candidate (2026-10-10 comment on perf(query): property-equality lookups and expansions cost O(dataset) per query on durable projects #1931):
GraphForge UUID build nominationat 478 MiB plus 30 MiB ofHashJoinInputexhausted the pool.
A read that a 4 GiB envelope could afford fails because of an arbitrary constant, and because the build side of a join has no path to disk.
Expected
- The query budget derives from the process's memory limit (cgroup
memory.max, then host memory) with a stated fraction and floor, the way the bulk builder already reads its cgroup budget; an explicitGraphForgeOptionsbudget still overrides it. - Operators that can spill (sort, aggregate) spill to the configured spill directory under the existing cap when the budget is exceeded, rather than failing.
- Hash joins either build the smaller side (planner chooses build side by estimated cardinality, with the anchored side as the probe) or partition the build side so each partition fits; a join that still cannot fit returns the typed
GF_EXECUTIONresource error it returns today, with the consumers named.
Acceptance criteria
- IC3's three pinned bindings complete on the SF1 SNB Interactive project under the 4 GiB envelope with correct results; latencies posted.
- A facade test runs a join whose build side exceeds a deliberately small budget and either spills or re-sides, with the result unchanged, and the pool's peak reservation asserted below the budget.
- Budget derivation is covered by a test with a fake cgroup
memory.max. - No test or production code sets a budget just to make a test pass.
Activity
DecisionNerd commentedon Oct 12, 2026
Widened by the probe snapshot on #1982 (2026-10-12): the fixed pool also fails aggregates, not only joins. IC6 fails
Resources exhausted … GroupedHashAggregateStream, IC9 fails… AggregateStream, IC7 failsHashJoinInput. Aggregates can spill in DataFusion but spill is disabled by default (spill_enabled: false), so they fail at 512 MiB too. Scope here covers both: derive the budget from the process limit, enable bounded spill for sort and aggregate, and re-side or partition hash joins. Acceptance adds IC6, IC7 and IC9 pinned bindings completing under the 4 GiB envelope with correct results.feat(exec): spill by fair share, plan unfittable hash joins as sort-m…
feat(api): derive the query memory budget from the process memory lim…
feat(exec): spill by fair share, plan unfittable hash joins as sort-m…
fix(exec): keep the greedy query pool; size spill tests by the partit…
test(api): report the derived query budget; drop the oversized-batch …
feat(api): derive the query memory budget from the process memory lim…
feat(exec): spill by fair share, plan unfittable hash joins as sort-m…
fix(exec): keep the greedy query pool; size spill tests by the partit…
44 remaining items
test(api): gate an anchored traversal's hash join and pruned enrichme…
fix(storage): bound an equality-index scan's row estimate by the inde…
test(api): gate an anchored traversal's hash join and pruned enrichme…
DecisionNerd commentedon Oct 13, 2026
U4 status (PR #2049,
perf/1985-anchored-varlen-build-estimate, head e6b4b4b).Fix: a current persisted equality index now bounds
PropertyOverlayExec's published row count by the manifest shards that can hold the value, so an anchored seed is no longer estimated as 20% of the route (IS6 traced at 636k seed rows, truth 1, join-left estimate 1,420,322,040 B against the 1 GiB limit under a 4G scope). The flip threshold andexpand_statistics.rsare unchanged.Verified:
graphforge-exec1131/1131 passed; the neighbouringgraphforge-apisuites (includingquery_memory_budget,traversal_join_memory,enrichment_frontier_reads,variable_length_frontier_reads,partitioned_expand,fixed_hop_limit) 91/91 passed, 2 skipped (release-only ignored benches); bdd 118/118 and TCK 3898/3898 with 0 regressed;make checkexit 0. The new facade test fails on revert at the plan assertion (SortMergeJoinExec plus an unpruned enrichment).NOT measured: IS6 / IS2 / IC2 / IC9 under
MemoryMax=4Gon the SF1 IC copy. I stopped before the release builds becausedf /fell below the 40 GB floor (24 GB free, 98% used, during the run) and the host load was 31. The p50 < 1 s target is therefore unconfirmed. When measuring, rebuild the property indexes on the copy with the same build as the binary under test (#2044 changed the pack format), and build the "before" binary from the same tree withproperty_scan.rsandproperty_equality_index.rsat theirmainversions.DecisionNerd commentedon Oct 13, 2026
PR #2050 (Part of #1985) fixes the TCR2 CPU regression from #2020: the partitions queued on the shared pool mutex through per-path-state
try_grow/shrink, then on the decoded CSR shard cache mutex.TCR2 SF1, bindings b0-b2 combined, median of 3 rounds (CPU-seconds,
systemd-run --scope -p MemoryMax=4G)Regression checks (median CPU-s): TCR1 2.6 -> 2.8 (ab6), 3.1 -> 3.1 (ab8); TCR8 14.5 -> 14.7 (ab6), 16.7 -> 16.1 (ab8, load 22-28). All bindings returned the reference row counts.
rebuild_property_indexes, needed after perf(storage): write edge-property pack columns as small authenticated pages (Part of #1984) #2044). Every arm read that same copy.Raw per-round results (ab6, ab7, ab8)
ab6
ab7
ab8
DecisionNerd commentedon Oct 13, 2026
U4 measurement on a real SF1 IC copy (PR #2049, head e6b4b4b)
Result: real IS6 no longer flips at the 4G-derived budget. p50 3.74 s -> 0.22 s under
MemoryMax=4G. IS2, IC2 and IC9 are unchanged.Method
cp -acopies of the SF1 IC project (targetedchmod u+w, originals untouched), one per build.e6b4b4b8d(based ond2f07c2b7, includes perf(storage): write edge-property pack columns as small authenticated pages (Part of #1984) #2044). Before = the same tree with onlyproperty_scan.rsandproperty_equality_index.rschecked out fromorigin/main(identical tod2f07c2b7for those files).rebuild_property_indexeswas run on each copy by the same build as the binary under test (perf(storage): write edge-property pack columns as small authenticated pages (Part of #1984) #2044 changed the pack format): 778 equality files / 77,560,132 B and 40 edge-property files / 159,919,725 B on both copies.systemd-run --user --scope -q -p MemoryMax=4G -p MemorySwapMax=0,run_suite.pybindings 0-2, two rounds interleaved before/after (round 2 reversed order), so 6 samples per row. Wall is per binding; CPU is user+sys of the whole probe process (3 bindings plus the ~0.35 s project open) from/usr/bin/time.Rows: identical between builds for every binding. Against the earlier reference run (
gdc-diag-1914b/results/ic*.json), IC2 (20,20,20) and IS2 (6,10,10) match. IS6 returns 1 row per binding (one container forum and moderator per message). IC9 errored in that old reference run, so its 20 rows are compared before vs after only.Plan check (
explainparam, IS6 bindings 0-2, run inside the same 4G scope): before has 2SortMergeJoinExec(both Left enrichment joins,on=[(node_uuid, node_uuid)]); after has 0SortMergeJoinExecand 8HashJoinExecon every binding. So the sort-merge left-enrichment pruning noted in the issue is not needed for this shape and is not added.Still open and not part of this change: IS2 (0.8 s) and IC2 (2.0 s) are under contended load and unchanged by the PR; IC9 (18 s, 145 s CPU) is unchanged and belongs to its own unit.
fix(exec): plan a join whose build side may outgrow the pool as a sor…
DecisionNerd commentedon Oct 14, 2026
Fresh SF1 IC3 proof at
fbf5a3c63exposed one remaining join reservation defect, so this PR stays draft despite green exact-head CI. All three pinned bindings were attempted once after an excluded warmup: p000 took 3.199 s (1 row), p002 took 2.742 s (0 rows), both exact reference/digest matches; p001 returnedGF_EXECUTIONwhen its empty external sorter requested 35.3 MB against 34.0 MB heuristic quota. The actual pool held 544.4 MB of a 2 GiB budget.The native process-tree highwater was 2,523,705,344 bytes under observed
memory.max=4294967296, with no OOM or process swap. Catalog counts reconciled. This is 2 matched / 3 attempted, 1 failed, 0 unrun, not a passing variant median or full coverage claim. No failed binding was retried. Raw driver/results/reference/native evidence and command/hash manifest preserve the failure.The cause is verified in DataFusion: an empty sorter cannot spill, so refusal of its first batch is fatal. Two new deterministic regressions reproduce that refusal before the repair, including an actual external sort. The bounded repair will apply producer headroom once a sorter has buffered input; its first batch still must fit the real pool and aggregate floors. No budget, timeout, reference, or assertion is being relaxed. A changed-source validation and fresh CI will be required before merge.
fix(exec): admit empty sort buffers against the real pool
DecisionNerd commentedon Oct 14, 2026
Fresh changed-source IC3 validation passed at
d77d09f895a89d9bbe4a2413e237f28f678089df. All three pinned SF1 bindings completed once after an excluded warmup, under the observed native 4 GiB process-tree limit and 120 s per-binding timeout.Nearest-rank p50 2.989806018 s, p95 3.190587422 s. 3 checked / 3 matched, 0 unchecked, 0 failures or unrun selected bindings. Driver/checker exit 0; total/per-label/per-type counts reconciled. Native highwater 2,559,234,048 bytes; no OOM or process swap. The reference is the harness's spec-derived IC reference; this is IC3-only diagnostic proof (3 of the full Interactive suite's 600 bindings), not audited certification or the parent-suite gate.
Raw results, reference, native recordings, exact commands and all input/output/executable hashes. The previous FBF first-buffer failure remains intact; no failed binding was retried on unchanged source. Two new regressions fail against that old rule and pass with the bounded repair; the real sort verifies 262,144 exact sorted values, actual spilling, the original real cap and reservation cleanup.
All 58 focused runtime/golden/facade tests passed, including the original 128 MiB join tests and 16 MiB extremum controls. Fake-cgroup budget derivation remains covered by the existing resource-policy tests. Final make check, formatting and diff checks passed. Independent final review is clear; CodeRabbit's two-file repair review completed with zero findings. Current-head CI and fresh GitHub reviews remain pending; PR #2115 is ready for review and will merge only after the required gate and review resolution.
fix(exec): bound sizing work to the next compact sort run
DecisionNerd commentedon Oct 14, 2026
Final current-head IC3 proof passed at
c6bb3c5f6db3a31991f513239375a3edc8dd8b8e: 3/3 pinned bindings completed and matched the exact spec-derived harness reference and emitted digests, with the excluded warmup completed. Driver latencies p000/p001/p002: 3.174785956 / 2.968538840 / 2.580869662 seconds; nearest-rank p50 2.968538840 s. Native process-tree highwater 2,647,412,736 bytes, actual 4 GiB memory limit, no OOM or process swap, per-binding timeout 120 s. Driver and reference checker exited 0; catalog counts reconciled. This is IC3-only diagnostic proof (3/600 Interactive bindings), not the full parent baseline or audited certification.All 59 focused runtime/golden/facade tests, final
make check, formatting and diff checks passed. The verified Greptile view-sizing finding is fixed and its thread resolved; independent current-source review is clear and CodeRabbit's two-file review completed with zero findings. Exact-head CI Gate passed with all product lanes green. PR #2115 is CLEAN and closes exactly this issue. It is ready for the squash merge queue.The earlier failed FBF diagnostic and D77 passing diagnostic remain preserved. No failed binding was retried on unchanged source; this new validation follows the reviewed compact-window repair.
Final C6 raw driver, input/executable hashes, native evidence and emitted result JSON
ic3-evidence.json
{ "schema": "graphforge-gdc-query-evidence/1", "certification": false, "suite": "snb-interactive", "status": "passed", "query_timeout_seconds": 120, "failures": [], "driver": { "name": "graphforge-benchmark-gdc-scorecard query", "version": "0.0.0", "executable_sha256": "2cfcdaa6cccc2df5397507fc127d9cf16471c5050145e8fb8bfc671f500e59da" }, "inputs": { "workload_sha256": "d0ad62c9ef96260bdb4ed224e0c5ba0b75ad9905cc12e7b6f764be1b4cd0ef57", "expected_counts_sha256": "9daa4440bfbac2f7d3b8165445f4afab67890cb71267d85a7691680269926feb" }, "project": { "path": "/home/ubuntu/gf-beta-evidence/join-candidate-view-window-fixed/ic", "opened_with": "graphforge_api::GraphForge::new(Some(path))" }, "reconciliation": { "status": "reconciled", "source": "https://github.com/ldbc/ldbc_snb_docs/blob/b2269610f433da72e7c97041f01680aae369a903/tables/legacy/table-number-of-entities-interactive.tex (sf1; every table reconciles to the listed records of the pinned archive)", "method": "Cypher count probes through GraphForge::execute, plus GraphForge::labels", "nodes": { "expected": 3181724, "observed": 3181724 }, "edges": { "expected": 17256038, "observed": 17256038 }, "labels": { "City": { "expected": 1343, "observed": 1343 }, "Comment": { "expected": 2052169, "observed": 2052169 }, "Company": { "expected": 1575, "observed": 1575 }, "Continent": { "expected": 6, "observed": 6 }, "Country": { "expected": 111, "observed": 111 }, "Forum": { "expected": 90492, "observed": 90492 }, "Person": { "expected": 9892, "observed": 9892 }, "Post": { "expected": 1003605, "observed": 1003605 }, "Tag": { "expected": 16080, "observed": 16080 }, "TagClass": { "expected": 71, "observed": 71 }, "University": { "expected": 6380, "observed": 6380 } }, "types": { "CONTAINER_OF": { "expected": 1003605, "observed": 1003605 }, "HAS_CREATOR": { "expected": 3055774, "observed": 3055774 }, "HAS_INTEREST": { "expected": 229166, "observed": 229166 }, "HAS_MEMBER": { "expected": 1611869, "observed": 1611869 }, "HAS_MODERATOR": { "expected": 90492, "observed": 90492 }, "HAS_TAG": { "expected": 3721417, "observed": 3721417 }, "HAS_TYPE": { "expected": 16080, "observed": 16080 }, "IS_LOCATED_IN": { "expected": 3073621, "observed": 3073621 }, "IS_PART_OF": { "expected": 1454, "observed": 1454 }, "IS_SUBCLASS_OF": { "expected": 70, "observed": 70 }, "KNOWS": { "expected": 180623, "observed": 180623 }, "LIKES": { "expected": 2190095, "observed": 2190095 }, "REPLY_OF": { "expected": 2052169, "observed": 2052169 }, "STUDY_AT": { "expected": 7949, "observed": 7949 }, "WORK_AT": { "expected": 21654, "observed": 21654 } } }, "latency_clock": { "id": "graphforge-gdc-query-clock/1", "producer": "graphforge-benchmark-gdc-scorecard query", "source": "std::time::Instant", "unit": "ns", "interval": "one public GraphForge API call returning a fully materialized Arrow result", "warmup_passes_per_variant": 1, "percentile_method": "nearest_rank" }, "result_digest": "graphforge-gdc-result-digest/1", "variants": [ { "query_id": "IC3", "interface": "graphforge_api::GraphForge::execute_with_params", "ordered": true, "columns": [ "friendId", "friendFirstName", "friendLastName", "xCount", "yCount", "xyCount" ], "status": "measured", "warmup": { "binding_id": "p000", "excluded": true, "completed": true }, "samples": [ { "binding_id": "p000", "status": "measured", "latency_ns": 3174785956, "rows": 1, "result_sha256": "39fe4a987576dc1620e3aaad2b700fbdadc77f5cc67e9523e38bbe251afadf69" }, { "binding_id": "p001", "status": "measured", "latency_ns": 2968538840, "rows": 0, "result_sha256": "0c50a11692c6f222c4fa3c454f546946bd4ac403cd1be936d74e18b2859cbfcd" }, { "binding_id": "p002", "status": "measured", "latency_ns": 2580869662, "rows": 0, "result_sha256": "0c50a11692c6f222c4fa3c454f546946bd4ac403cd1be936d74e18b2859cbfcd" } ], "summary": { "count": 3, "p50_ns": 2968538840, "p95_ns": 3174785956 } } ] }ic3-checked-report.json
{ "scope": "diagnostic; does not certify full pinned coverage", "source_sha": "c6bb3c5f6db3a31991f513239375a3edc8dd8b8e", "driver_sha256": "2cfcdaa6cccc2df5397507fc127d9cf16471c5050145e8fb8bfc671f500e59da", "driver_evidence_sha256": "99dac60a5cdc260e69779b5b47ed73edb3499b98ef53e4fcd646476d2eaf75c5", "native_ok": true, "native": { "starttime": "2026-10-14 17:24:11.763720+00:00", "walltime": 19.253061415976845, "cputime": 59.779501, "memory": 2647412736, "blkio-read": 807706624, "blkio-write": 3290451968, "pressure-cpu-some": 0.098013, "pressure-memory-some": 0.0, "pressure-io-some": 0.174139, "exitcode": "return value 0", "exit_value": 0, "exit_signal": null }, "variants": [ { "query_id": "IC3", "status": "measured", "samples": 3, "measured": 3, "failures": [], "latencies_seconds": [ 3.174785956, 2.96853884, 2.580869662 ], "p50_seconds": 2.96853884, "warmup": { "binding_id": "p000", "excluded": true, "completed": true } } ], "correctness": { "schema": "graphforge-gdc-rung-correctness/1", "status": "passed", "reference_source": "a spec-derived reference (gdc_snb_interactive_reference over the loaded snapshot), not the LDBC validation set (the pinned validation stream begins with an update at position 0)", "reference_sha256": "723cde5975f11f290fc6ad342b8dfdb7172a3cc18b85de1200ebcc5cd3d7010f", "checked": 3, "matched": 3, "unchecked": 0, "queries": { "IC3": { "matching": "exact", "checked": 3, "matched": 3 } }, "mismatches": [] }, "all_selected_bindings_completed": true, "all_selected_p50_at_most_30_seconds": true }execution-inputs.json
{ "source_sha": "c6bb3c5f6db3a31991f513239375a3edc8dd8b8e", "source_clean": true, "changed_source_validation_after_verified_view_window_fix": true, "sha256": { "/home/ubuntu/gf-beta-evidence/join-prepared-view-window-fixed/rebuild-copy-tool/src/main.rs": "179a2c6a9d663a6e64969e7f717e2ab0c1ad2095264dc14de6725dc5b395d288", "/home/ubuntu/gf-beta-evidence/join-prepared-view-window-fixed/rebuild-copy-tool/Cargo.toml": "00379a66fe591614964ecc7f7b96c487a9c9e6f2b8ff9d9138d509b65698ffb5", "/home/ubuntu/gf-beta-evidence/join-prepared-view-window-fixed/rebuild-copy-tool/Cargo.lock": "7c5b14ca224d8cb4e21a01f67ce1fe16cad708e9fc2a91c09a194f61a9a87e60", "/home/ubuntu/gf-beta-evidence/join-prepared-view-window-fixed/input-sha256-manifest.json": "23f8fdc0c4cb87bb4aadbc321702fc299e6131fae1ba40b598be6df549888092", "/home/ubuntu/gf-beta-scorecard-target/release/gf-beta-rebuild-copy": "631b445698c91ebf7a12924256e216b4390eb19bc4ded86f2c0371f80759fb56", "/home/ubuntu/gf-beta-scorecard-target/release/graphforge-benchmark-gdc-scorecard": "2cfcdaa6cccc2df5397507fc127d9cf16471c5050145e8fb8bfc671f500e59da", "/home/ubuntu/gf-beta-scorecard-target/release/gf": "d0a90f7269668b78f4567be313cb731f407a983060008cf095c8fea3486fd7ef", "/home/ubuntu/gf-beta-evidence/join-candidate-view-window-fixed/ic3-workload.json": "d0ad62c9ef96260bdb4ed224e0c5ba0b75ad9905cc12e7b6f764be1b4cd0ef57", "/home/ubuntu/gf-beta-evidence/join-candidate-view-window-fixed/ic3-reference.json": "723cde5975f11f290fc6ad342b8dfdb7172a3cc18b85de1200ebcc5cd3d7010f", "/home/ubuntu/gf-beta-evidence/join-prepared-view-window-fixed/snb-interactive/expected-counts.json": "9daa4440bfbac2f7d3b8165445f4afab67890cb71267d85a7691680269926feb" } }native-ic3/invocation.json
{ "source": "/home/ubuntu/gf-join-finish", "source_sha": "c6bb3c5f6db3a31991f513239375a3edc8dd8b8e", "command": [ "/home/ubuntu/gf-beta-scorecard-target/release/graphforge-benchmark-gdc-scorecard", "query", "--project", "/home/ubuntu/gf-beta-evidence/join-candidate-view-window-fixed/ic", "--workload", "/home/ubuntu/gf-beta-evidence/join-candidate-view-window-fixed/ic3-workload.json", "--expected-counts", "/home/ubuntu/gf-beta-evidence/join-prepared-view-window-fixed/snb-interactive/expected-counts.json", "--output", "/home/ubuntu/gf-beta-evidence/join-candidate-view-window-fixed/ic3-evidence.json", "--results-dir", "/home/ubuntu/gf-beta-evidence/join-candidate-view-window-fixed/ic3-results", "--query-timeout-seconds", "120" ], "command_sha256": "2cfcdaa6cccc2df5397507fc127d9cf16471c5050145e8fb8bfc671f500e59da", "wrapper_sha256": "97e0cae501b3f49ae49f62a021c208c3555eedb7201f2181161817a30c3590a5", "phase_helper_sha256": "50240560bee83ead3208c62e6ffda0993014dad05495060e30453fbe4aa6a46b", "benchexec_version": "3.35", "python": "/usr/bin/python3", "python_sha256": "52e0a13e60a981d8c4b6478be2ba5176f69da07948a056bf49cf6f077e30cb41", "memlimit": 4294967296, "walltimelimit": 14400, "working_dir": "/home/ubuntu/gf-beta-evidence/join-candidate-view-window-fixed", "output": "/home/ubuntu/gf-beta-evidence/join-candidate-view-window-fixed/native-ic3", "writable_project": "/home/ubuntu/gf-beta-evidence/join-candidate-view-window-fixed", "container": true, "root_mode": "read-only", "container_tmpfs": false, "environment": { "PATH": "/usr/bin:/bin", "LANG": "C.UTF-8", "LC_ALL": "C.UTF-8", "TMPDIR": "/home/ubuntu/gf-beta-evidence/join-candidate-view-window-fixed/native-ic3/tmp" }, "host_filesystem": "{\n \"filesystems\": [\n {\n \"target\": \"/\",\n \"source\": \"/dev/md3\",\n \"fstype\": \"ext4\",\n \"options\": \"rw,relatime\"\n }\n ]\n}\n" }native-ic3/runexecutor-result.json
{ "starttime": "2026-10-14 17:24:11.763720+00:00", "walltime": 19.253061415976845, "cputime": 59.779501, "memory": 2647412736, "blkio-read": 807706624, "blkio-write": 3290451968, "pressure-cpu-some": 0.098013, "pressure-memory-some": 0.0, "pressure-io-some": 0.174139, "exitcode": "return value 0", "exit_value": 0, "exit_signal": null }native-ic3/native-cgroup-facts.json
{ "runs": { "/sys/fs/cgroup/user.slice/user-1000.slice/[email protected]/benchexec.slice/run-p2052196-i39826230.scope/benchmark_4m_qzbdh": { "cgroup_path": "/sys/fs/cgroup/user.slice/user-1000.slice/[email protected]/benchexec.slice/run-p2052196-i39826230.scope/benchmark_4m_qzbdh", "memory.max": "4294967296", "memory.swap.max": "0", "cgroup.procs": "", "memory.events": "low 0\nhigh 0\nmax 0\noom 0\noom_kill 0\noom_group_kill 0\nsock_throttled 0", "observed_pids": [] } }, "observer_errors": [], "requested_memory_bytes": 4294967296, "observed_4GiB": true, "swap_observed": true, "swapped": false }ic3-workload.json
{ "schema": "graphforge-gdc-query-workload/1", "suite": "snb-interactive", "variants": [ { "bindings": [ { "id": "p000", "params": { "countryXName": { "type": "Str", "value": "Laos" }, "countryYName": { "type": "Str", "value": "Scotland" }, "endDate": { "type": "Int", "value": 1310515200000 }, "personId": { "type": "Int", "value": 17592186055119 }, "startDate": { "type": "Int", "value": 1306886400000 } } }, { "id": "p001", "params": { "countryXName": { "type": "Str", "value": "Uruguay" }, "countryYName": { "type": "Str", "value": "Scotland" }, "endDate": { "type": "Int", "value": 1301356800000 }, "personId": { "type": "Int", "value": 8796093030404 }, "startDate": { "type": "Int", "value": 1298937600000 } } }, { "id": "p002", "params": { "countryXName": { "type": "Str", "value": "Scotland" }, "countryYName": { "type": "Str", "value": "Slovakia" }, "endDate": { "type": "Int", "value": 1304208000000 }, "personId": { "type": "Int", "value": 2199023266354 }, "startDate": { "type": "Int", "value": 1301616000000 } } } ], "id": "IC3", "operation": { "kind": "cypher", "text": "MATCH (countryX:Country {name: $countryXName}),\n (countryY:Country {name: $countryYName}),\n (person:Person {id: $personId})\nWITH person, countryX, countryY\nLIMIT 1\nMATCH (city:City)-[:IS_PART_OF]->(country:Country)\nWHERE country IN [countryX, countryY]\nWITH person, countryX, countryY, collect(city) AS cities\nMATCH (person)-[:KNOWS*1..2]-(friend)-[:IS_LOCATED_IN]->(city)\nWHERE NOT person = friend AND NOT city IN cities\nWITH DISTINCT friend, countryX, countryY\nMATCH (friend)<-[:HAS_CREATOR]-(message),\n (message)-[:IS_LOCATED_IN]->(country)\nWHERE $endDate > message.creationDate >= $startDate AND\n country IN [countryX, countryY]\nWITH friend,\n CASE WHEN country = countryX THEN 1 ELSE 0 END AS messageX,\n CASE WHEN country = countryY THEN 1 ELSE 0 END AS messageY\nWITH friend, sum(messageX) AS xCount, sum(messageY) AS yCount\nWHERE xCount > 0 AND yCount > 0\nRETURN friend.id AS friendId,\n friend.firstName AS friendFirstName,\n friend.lastName AS friendLastName,\n xCount,\n yCount,\n xCount + yCount AS xyCount\nORDER BY xyCount DESC, friendId ASC\nLIMIT 20" }, "ordered": true, "columns": [ "friendId", "friendFirstName", "friendLastName", "xCount", "yCount", "xyCount" ] } ] }ic3-reference.json
{ "queries": { "IC3": { "bindings": { "p000": { "columns": [ "friendId", "friendFirstName", "friendLastName", "xCount", "yCount", "xyCount" ], "rows": [ [ "8796093029689", "Eun-Hye", "Yoon", "1", "1", "2" ] ] }, "p001": { "columns": [ "friendId", "friendFirstName", "friendLastName", "xCount", "yCount", "xyCount" ], "rows": [] }, "p002": { "columns": [ "friendId", "friendFirstName", "friendLastName", "xCount", "yCount", "xyCount" ], "rows": [] } }, "matching": "exact" } }, "rung_id": "sf1", "schema": "graphforge-gdc-rung-reference/1", "source": "a spec-derived reference (gdc_snb_interactive_reference over the loaded snapshot), not the LDBC validation set (the pinned validation stream begins with an update at position 0)", "suite_id": "snb-interactive" }00000000.json
{"binding_id":"p000","columns":[{"name":"friendId","type":"Int64"},{"name":"friendFirstName","type":"Utf8"},{"name":"friendLastName","type":"Utf8"},{"name":"xCount","type":"Int64"},{"name":"yCount","type":"Int64"},{"name":"xyCount","type":"Int64"}],"ordered":true,"query_id":"IC3","result_sha256":"39fe4a987576dc1620e3aaad2b700fbdadc77f5cc67e9523e38bbe251afadf69","rows":[["8796093029689","Eun-Hye","Yoon","1","1","2"]],"schema":"graphforge-gdc-query-result/1"}00000001.json
{"binding_id":"p001","columns":[{"name":"friendId","type":"Int64"},{"name":"friendFirstName","type":"Utf8"},{"name":"friendLastName","type":"Utf8"},{"name":"xCount","type":"Int64"},{"name":"yCount","type":"Int64"},{"name":"xyCount","type":"Int64"}],"ordered":true,"query_id":"IC3","result_sha256":"0c50a11692c6f222c4fa3c454f546946bd4ac403cd1be936d74e18b2859cbfcd","rows":[],"schema":"graphforge-gdc-query-result/1"}00000002.json
{"binding_id":"p002","columns":[{"name":"friendId","type":"Int64"},{"name":"friendFirstName","type":"Utf8"},{"name":"friendLastName","type":"Utf8"},{"name":"xCount","type":"Int64"},{"name":"yCount","type":"Int64"},{"name":"xyCount","type":"Int64"}],"ordered":true,"query_id":"IC3","result_sha256":"0c50a11692c6f222c4fa3c454f546946bd4ac403cd1be936d74e18b2859cbfcd","rows":[],"schema":"graphforge-gdc-query-result/1"}DecisionNerd commentedon Oct 14, 2026
The complete final C6 IC3 diagnostic bundle is now published: https://gist.github.com/DecisionNerd/461fa55cf5b1aef762584b7a4a45db01 . All 34 nonempty files were verified byte-for-byte after publication; the manifest records three omitted empty files. It includes native cgroup records, all three emitted results and references, exact input/binary hashes, the 59-test/check logs, completed CodeRabbit review, and the old-code regression reproduction.
Coverage remains IC3’s first three bindings only (3/600 Interactive bindings), with all three reference matches and certification=false. This is the final reviewed candidate’s evidence; full-suite beta coverage remains outstanding on #1982. The bundle preserves the queue state observed at staging and makes no merge claim.