Gleb Sinyavskiy created TINKERPOP-2855:
------------------------------------------
Summary: Performance degradation in TinkerGraph 3.5.4
Key: TINKERPOP-2855
URL: https://issues.apache.org/jira/browse/TINKERPOP-2855
Project: TinkerPop
Issue Type: Bug
Components: tinkergraph
Affects Versions: 3.5.4
Environment: Ubuntu 22.04.1, docker
Reporter: Gleb Sinyavskiy
Hello,
I few days ago I tried to update gremlin-server in our project from 3.5.3 to
3.5.4 and found out that it's test suite got 5 times slower. We use the
official docker image(tinkerpop/gremlin-server) with some configuration changes:
{code:java}
gremlin.tinkergraph.vertexIdManager=ANY
gremlin.tinkergraph.edgeIdManager=ANY {code}
The app uses user-generated string ids and heavily relies on the upsert pattern
from the [recipes|https://tinkerpop.apache.org/docs/3.5.4/recipes/].
Yesterday I made an investigation and narrowed it down to the performance on
the upsert pattern. I also discovered that the issue is not related to our
configuration changes and can be reproduced with vanilla image and LONG ids.
I prepared a [simple script that reproduces the
issue|https://github.com/zhulik/gremlin-server-performance-issue/] and
contacted Stephen Mallette on discord. They confirmed the issue exists, but
only in 3.5.4, 3.6.0 performs as expected. They also wrote a groovy script that
reproduces the problem:
{code:groovy}
g = TinkerGraph.open().traversal()
batches = (0..<100).collect{ (0..<100) }
start = System.currentTimeMillis()
for (batch in batches) {
b = g
for (id in batch) {
b = b.V(id).fold().coalesce(__.unfold(), __.addV("test").property(T.id, id))
}
b.iterate()
}
System.currentTimeMillis() - start
{code}
[discord
message|https://discord.com/channels/838910279550238720/838910279550238723/1064964247823593502]
Both my and Stephen's scripts perform few times slower when executed against
3.5.4 in compare to 3.5.3 or 3.6.0.
*Steps to reproduce:*
Run my or Stephen's script against vanilla tinkerpop/gremlin-server:3.5.4
docker image
*Expected result:*
Script's execution time matches it's time executed against 3.5.3 or 3.6.0
*Observed result:*
Script is 5-7 times slower.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
