Storaged memory continues to grow until the leader change or OOM

[carrey-feng]

Describe the bug (required)
All storageds memory continues to grow until someone OOM or leader changes by running balance leader command when leader is unblanced in cluster.
I used jemalloc to dump malloc, I found many "nebula::ConstantExpression::clone" before OOM. and it disappeared or became less after OOM or Leader changed.

I found that "nebula::ConstantExpression::clone" was called by fetching default value of edge property.
I do not know if it leaked and why it became less after Leader changed.
Can you give me some ideas on how to check. thanks

Your Environments (required)

• OS: Centos 7.5
• Compiler: g++ --version : 4.8.5
• CPU: lscpu
• Version: 2.6.1
• Commit id : de03025 with pr add session tool to manage and analysis sessions and Limit number of sessions that are created by same client ip and same user for each graphd

How To Reproduce(required)

Steps to reproduce the behavior:

1. Step 1
2. Step 2
3. Step 3

Expected behavior

Additional context

08:35 jeprof top

08:36 jeprof top

08:37 jeprof top

08:35 call graph
210835.pdf

[cangfengzhs]

Yes, this is a serious bug.And you create a tag/edge with a default value, right? We have found and fixed it. Look at #3601.

[carrey-feng]

Yes, this is a serious bug.And you create a tag/edge with a default value, right? We have found and fixed it. Look at #3601.

It is great ! I have deleted edge default value, memory usage is very flat. Thanks!

Another problem that #3729 was create at last week, there is anyone to review it so far. I think that it is import to multi-tenant cluster. It can prevent one user's mistake from affecting all other users. Could you review it ? Thanks !

[Sophie-Xie]

Yes, this is a serious bug.And you create a tag/edge with a default value, right? We have found and fixed it. Look at #3601.

It is great ! I have deleted edge default value, memory usage is very flat. Thanks!

Another problem that #3729 was create at last week, there is anyone to review it so far. I think that it is import to multi-tenant cluster. It can prevent one user's mistake from affecting all other users. Could you review it ? Thanks !

Thanks. I just assigned two reviewers.

[wey-gu]

Thanks @a516072575 , great option introduced.