imbajin opened a new pull request, #835: URL: https://github.com/apache/hugegraph-toolchain/pull/835
## Purpose of the PR Publish the prioritized optimization TODO, source-level analysis and reproducible local Hubble + PD + HStore measurements. This PR contains documentation, benchmark scripts and evidence only. ## Main Changes - Preserve the original 1/10/100/500/1,000 graphspace scaling study, 50-graph request breakdown, direct PD RPC baseline and 12,000-vertex/12,000-edge datasets with 64 B and 8 KiB properties. - Add a controlled before/after comparison using identity encoding and the same Server build; record sample counts, response bytes, P50/P95, CPU profiles, browser screenshots and provenance. - Verify default 300-row reads, 1,000-node previews, zero-POST large-query preflight, existing async execution and a separate 100 KiB/1 MiB attribute fixture with original-value SHA checks and read-only/import safeguards. - Include the final shared 10,000-value property budget, local reducer/explicit mapper bound checks, and negative tests for proxy errors, async timeouts and missing fixture edges. - Add final review-boundary API assertions and a fully isolated PD authentication fixture: real administrator and ordinary-member Chromium login, query, permission visibility and logout; preserve original-data SHA checks and bounded-error evidence. - Record the separate Server compression fix and its unchanged 16 MiB result-limit rejection, with a licensed Java reproduction harness and exact commands. - Keep regressions, partial-preview semantics, baseline-anonymous performance measurements and separate real-role browser coverage and remaining O(N) metadata work explicit. No runtime credentials, configurations, databases or distribution JARs are included. ## Results and related changes - Graph-list page P50: 267.73 → 35.70 ms (one graph), 230.88 → 87.79 ms (ten graphs). - Final navigation observations: 10.68 → 2.89 s / 12.79 → 3.33 s. These are browser observations, not P50. - 8 KiB, 1,000-vertex preview: 25.41 MB → 1.10 MB; large-property fixture: 4,509,209 → 9,924 bytes. - Same oversized async task reaches the same rejection in 126.471 → 1.600 s; these single observations are supported by a paired compression microbenchmark and are separate from the main Hubble timing window. - 1,000-node long tasks remain approximately 0.7–0.9 s, and some small-result cases do not improve. Implementation PRs: navigation #78, query/property previews #79, Server buffer growth https://github.com/hugegraph/hugegraph/pull/327. Start with [COMPARISON.md](https://github.com/hugegraph/hugegraph-toolchain/blob/ac1dbd1ef1d77e72378b98f2b10ffb2e7fd2fd5c/docs/performance/2026-10-22-hubble-hstore/COMPARISON.md) and [TODO.md](https://github.com/hugegraph/hugegraph-toolchain/blob/ac1dbd1ef1d77e72378b98f2b10ffb2e7fd2fd5c/docs/performance/2026-10-22-hubble-hstore/TODO.md); README and ANALYSIS-PLAN preserve the baseline and deeper analysis. Detailed local full-suite failures and focused reruns: [VALIDATION.md](https://github.com/hugegraph/hugegraph-toolchain/blob/ac1dbd1ef1d77e72378b98f2b10ffb2e7fd2fd5c/docs/performance/2026-10-22-hubble-hstore/VALIDATION.md). The report does not claim a single all-green frontend run. ## Verifying these changes - [x] Local source builds, related backend/frontend tests and Chromium checks recorded in the report. - [x] JSON/JSONL/gzip evidence files parsed; Python syntax, local report links, script license headers and credential-pattern scan passed. - [x] Docs-only diff, whitespace check and exact local/remote Git tree equality verified. ## Documentation Status - [x] Doc - Done Final review supplement: the report retains historical timings and adds the final 186-backend / 401-frontend passing checks, source-built hstore/Chromium evidence, bounded 30-minute Guava statistics caching, unknown/approximate count semantics, and mixed complete/partial property editing with actual wire IDs. These are functional checks, not fresh matched latency benchmarks. The report remains optional to merge. --- Source community PR: https://github.com/hugegraph/hugegraph-toolchain/pull/80. This upstream PR reuses the same branch. Figure PR numbers refer to that community source. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
