Summary:
This is done when the generated AST will be cached.
Remove LiteralsPlugger.
Reviewers: florijan, mislav.bradac
Reviewed By: mislav.bradac
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D788
Summary: I was mistaken in my calculations before and gave it +-3sigma tolerance (0.0027 probability of failure). Now I changed it to +-5sigma, which is good enough for CERN, and should be for us too.
Reviewers: mislav.bradac, mferencevic
Reviewed By: mislav.bradac
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D780
Summary:
Replaced std::list with std::vector in all plan operators. Performance increase in harness tests is not visible. Defined a custom test:
```
unwind range(0, 1000000) as x
create ({a: tointeger(rand() * 100), b: tointeger(rand() * 100), c: tointeger(rand() * 100), d: tointeger(rand() * 10), e: tointeger(rand() * 10), f: tointeger(rand() * 10)});
match (n) return min(n.a), max(n.b), sum(n.c), n.d, n.e, n.f
match (n) with distinct n.a AS a, n.b AS b, n.c AS c, n.d AS d, n.e AS e, n.f AS f return count(*)
```
In that test performance gains are 9.8% on the aggregation query (mean 0.83s vs 092s) and 34% (mean 2.15s vs 3.25s) on the distinct query. Doubt we'll see much on any of the LDBC tests because they don't stress those operators nearly as much.
Reviewers: buda, teon.banek, mislav.bradac
Reviewed By: teon.banek
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D778
Summary:
Antlr grammar has been updated to support putting edge types after the BFS
symbol. Planner collects edge type filters for BFS and inlines them in the
operator by joining the filter with the user input BFS filter itself. This
requires no change from the standpoint of the operator. On the other hand, in
order to use the faster lookup by a single edge type, `ExpandBreadthFirst`
operator now accept an optional edge type. The edge type is passed from the
planner only if the user is filtering by a single type.
Unit tests as well as tck have been updated.
Reviewers: florijan, mislav.bradac
Reviewed By: florijan
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D777
Summary: Reduces latency on LDBC query 9 from 7.9sec to 6.8sec (14%). That query has 650k rows in ORDER BY, 3 ordering elements and 10ish values get returned (both of them are now accumulated into vectors).
Reviewers: buda, mislav.bradac
Reviewed By: mislav.bradac
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D775
Summary:
Add function First to utils.
Insert EdgeType into Expand during planning.
Reviewers: florijan, mislav.bradac
Reviewed By: florijan
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D769
Summary: The constness of the DbAccessor interferes with caching the results.
Reviewers: florijan, mislav.bradac
Reviewed By: mislav.bradac
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D771
Summary:
Benchmark planning and estimating indexed ScanAll. According to the benchmark,
caching speeds up the whole process of planning and estimation by a factor of
2. Most of the performance gain is in the `CostEstimator` itself, due to plenty
of calls to `VerticesCount` when estimating all of the generated plans.
Reviewers: mislav.bradac, florijan
Reviewed By: mislav.bradac
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D765
Summary:
This fixes the problem when the test would fail on some PCs...
Notably, mine...
Reviewers: florijan, mferencevic, mislav.bradac, buda
Reviewed By: buda
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D770
Summary:
- The `Edges` data structure now handles common ops, including providing an iterator over edges whose "other" vertex is know.
- This should improve performance on dense_expand tests in the harness without other side-effects.
- query::plan::Expand operator modified not to check for existing-node stuff since that now gets handled by the `Edges` data structure.
- `Edges::Iterator` implemented only for const iterators since that suffices for now. Can implement non-const if the need arrises.
Reviewers: buda, mislav.bradac, teon.banek
Reviewed By: teon.banek
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D763
Summary: For the time being we'll go with vector and a custom iterator for know-destination-vertex lookups, but this benchmark might be handy to keep for future work.
Reviewers: buda, mislav.bradac, teon.banek
Reviewed By: mislav.bradac
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D762
Summary:
Measured improvements in a test scenario with 4 labels, 1M vertices for each label:
Old code: ~2.5 seconds per index
New code: ~1.5 seconds per index
When building an index for a non-existing label the updated code is done immediately. The old code depends on the number of vertices in the database.
The new code *could* be slower when building an index for a label that has a lot of vertices, and the index is not garbage collected recently and contains a lot of junk. This can be avoided by a simple check in the `BuildIndex` function (if label_index cardinality > total cardinality), if you like.
Reviewers: buda, mislav.bradac, teon.banek
Reviewed By: mislav.bradac
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D767
Summary: Ugly formatting and bad variable naming due to 1024 limit on neo4j client.
Reviewers: buda, teon.banek
Reviewed By: buda
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D748
Summary:
1. Test setup rewritten to take cca 8 seconds. Note that edges are created by using:
`MATCH (a) WITH a MATCH (b) WITH b WHERE rand() < X CREATE (a)-[:ET]->(b)`
Where `X` is a threshold calculated so the desired edge count is the expectation. This seems the only feasable way of generating a large number of edges since query execution does not depend on edge count, but on vertex count.
2. Using the new `assert` function to verify graph state. I recommend doing that in all the harness tests (I don't think we currently have something better).
3. All tests rewritten to take around 200ms per iteration.
4. Test are using SKIP to avoid sending data to the client, but ensure that appropriate operations get executed. This currently seems like the best way of removing unwanted side-effects.
Harness will cost us our sanity. And it doesn't even provide good quality regression testing we really need :(
Reviewers: buda, mislav.bradac, mferencevic
Reviewed By: mferencevic
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D752
Summary: It turns out trivial if I use unwind for vertex creation, MATCH for edge creation and UNWIND for test duration. It took hours to converge to this :(
Reviewers: mislav.bradac, buda
Reviewed By: mislav.bradac
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D747
Summary: `assert` function added. Useful to us for asserting DB state in harness tests. Potentially useful to the client for breaking out of a query as soon as a predicate fails, as opposed to collecting result and checking them client-side.
Reviewers: buda, mislav.bradac, teon.banek
Reviewed By: teon.banek
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D750
Summary:
Moved Neo4j config to config dir.
Neo4j and PostgreSQL are now downloaded to libs.
Renamed metadata flags in memgraph.
Changed apollo generate for new harness.
Reviewers: mislav.bradac
Reviewed By: mislav.bradac
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D741
