ProxyFactory memory leak using cglib [SPR-3620]

[spring-projects-issues]

Andy Piper opened SPR-3620 and commented

We are seeing a memory leak in ProxyFactory when using cglib. I will attach a picture of the references.

---

Affects: 2.0.5

Attachments:

• cglib.patch (7.25 kB)
• cglib.patch (6.18 kB)
• snap_cglib.jpg (190.08 kB)

Issue Links:

• OSGI-161 Class memory leak ("is duplicated by")
• MemoryLeak in Cglib2AopProxy.ProxyCallbackFilter [SPR-8008] #12663 MemoryLeak in Cglib2AopProxy.ProxyCallbackFilter
• Cglib2AopProxy weak reference causes perm space to fill up [SPR-4004] #8684 Cglib2AopProxy weak reference causes perm space to fill up

[spring-projects-issues]

Andy Piper commented

Memory reference picture

[spring-projects-issues]

Andy Piper commented

I think that the bug is in ProxyCallbackFilter. cglib maintains a hard reference to this in the key and so the value doesn't go away. I suspect by making the fixedInterceptorMap and advised fields weak this may cure the problem.

[spring-projects-issues]

Juergen Hoeller commented

This might be a general memory leak problem with CGLIB, I'm afraid. If we make Spring's fields weakly referenced there, the CGLIB class itself will probably still stay around in memory. Of course the leak would show less effect then... so that might already help.

Juergen

[spring-projects-issues]

Andy Piper commented

I'm not so sure. cglib correctly uses a WeakHashMap, with Spring fixed things might be ok. I'll attach the candidate fix that I am testing.

[spring-projects-issues]

Andy Piper commented

Candidate fix

[spring-projects-issues]

Andy Piper commented

This is confirmed to fix at least one leak.

[spring-projects-issues]

Andy Piper commented

Updated patch. I found another leak.

[spring-projects-issues]

Juergen Hoeller commented

I have fixed this in CVS HEAD (for 2.1 M3) and mbranch-2-0 (for 2.0.7). This should be available in the next 2.1 M3 as well as 2.0.7 snapshots (tonight / tomorrow).

Juergen

[spring-projects-issues]

Juergen Hoeller commented

Andy, could you please re-test a current 2.5 RC2 / 2.0.8 snapshot with respect to this leak? I had to revise Cglib2AopProxy again since the fixes that you suggested here unfortunately caused a side effect with the reuse of CGLIB proxy classes... (see #8684)

Thanks in advance,

Juergen