SebastianGruza commented on issue #3090:
URL: https://github.com/apache/hugegraph/issues/3090#issuecomment-5528339884
@contrueCT — we ran exactly that combined-branch test, as a black-box,
cross-backend suite: the same REST/Gremlin queries against two servers built
from the same tree, HStore (PD + 3 store nodes) vs RocksDB, compared as **sets
of element ids** rather than counts. Server = master `98477f0` + #3184
`1072872` + #3182 `acbee167` + #2994 `5cb9a519` (no merge conflicts; #3182 has
since been merged, so this equals today's master + #2994 + #3184). Same run on
plain master `98477f0` as the baseline.
| Your item | Section | Cases | master `98477f0` | combined |
|---|---|---|---|---|
| 1. #3184 sort-key prefix/range reproducer cases | S | 56 | 15 OK, **37
store-side decode errors** (#3090), 4 both-error | 52 identical sets, 4
both-error¹ |
| 2. #2994 single/multi/conflicting/IN/non-equality labels with edge sort
keys | J | 35 | 32 OK, 1 error (#3090 shape), **2 `without()` + range-index
queries throw on both backends** | 35 identical sets |
| 3. #3182 flattened `within()` + post-filtering, cache hit and miss | K |
40 | 24 OK, **16 wrong sets — HStore returns only the first `within()` value
when a SEARCH condition is present** (20 of 40/50/60/80, no error) | 40
identical, every query run twice, #3180 combo ×10 stable |
| 4. #3140 cross-partition limit/offset/page-cursor | L | 43 | 43 identical
| 43 identical across 3 partitions |
Version axis with the backend held constant (RocksDB master vs RocksDB
combined): 168/174 identical sets, the only differences being the two
`without()` cases that throw on master and return results on combined — i.e. no
other semantic change from #2994/#3182 visible to this suite.
¹ The 4 both-error cases are string range predicates in the REST
`properties` filter (`P.gte("ETC")` → `expect a number`) rejected on both
backends, while the same ranges work through Gremlin — a REST parser
limitation, not part of these PRs.
Suite, cluster scripts, exact branch recipe and raw reports (ids included,
so every mismatch can be re-derived):
https://github.com/SebastianGruza/hugegraph-oracle-suite — aggregated numbers
in `results/README.md`. It is Python 3 stdlib only and takes any two servers by
port, so it can be re-run against future heads of #2994/#3184. One unrelated
server-level finding from the same run (a duplicated record on page boundaries
when `limit` is a multiple of 500, identical on both backends) is filed
separately as #3191.
--
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]
