Settings for read optimized graph?

[porscheme]

General Question

Our graph is read only...

• We never do online UPSERTs
• We only update through SST files on weekly cadence

I wanted to now if there is any read optimized Nebula or RocksDB settings.
Currently graph walks are annoyingly very slow.
Below is our query...

MATCH (p:Student)
WHERE id(p) IN [ ... ]

OPTIONAL MATCH (p)<-[:HAS_COURSE]-(a:CourseCodes)
OPTIONAL MATCH (p)<-[:STUDENT_HAS_SOCIAL]-(s:Social)
OPTIONAL MATCH (p)<-[:HAS_PROCEDURE]-(r:ProcedureCodes)

WITH
  p,
  COLLECT(DISTINCT a.CourseCodes.CodeId) as courses,
  COLLECT(DISTINCT s.Social.AttributeId) as social,
  COLLECT(DISTINCT r.ProcedureCodes.CodeId) as procedures

RETURN 
  p.Student.StudentId as Student, 
  p.Student.Gender as sex, 
  p.Student.Race as race, 
  p.Student.Ethnicity as ethnicity, 
  p.Student.MaritalStatus as maritalstatus,
  diagnosis,
  procedures,
  social

[porscheme]

Anyone?

[Sophie-Xie]

@yixinglu Pls take a look, thanks.

[yixinglu]

sorry to reply late.

there some optimize options for performance tuning, u could try to update following flags separately:

1. nebula-graphd.conf

--optimize_appendvertices=true
--max_job_size=10

2. nebula-storaged.conf

--query_concurrently=true

In addition, you can profile above query in your environment to check where the bottleneck is at runtime.

by the way, what's the version of nebula you used? if possible, you could upgrade the latest version since we have improve the match performance in latest version.

[porscheme]

Our cluster configuration:
version: v3.4.0
metad: 3 (16 cores, 128 GB, 2 TB SSD)
graphd: 3 (16 cores, 128 GB, 2 TB SSD)
storaged: 5 (16 cores, 128 GB, 2 TB SSD)
Replica Factor: 3
VID: String 40 character size

We did see performance improvement after setting below but not enough. Is there anything else we can do?

• nebula-graphd.conf

--optimize_appendvertices=true
--max_job_size=10

• nebula-storaged.conf

# This is turned ON by default in nebula v3.4.0, just do it just incase
query_concurrently=true