Memory leak happening while using registerModule/unregisterModule.

[skkart]

Version

3.1.0

Reproduction link

https://jsfiddle.net/mrj8spu0/

Steps to reproduce

1. Open the JSFiddle link mentioned as part of this issue.
2. Check and record the heap memory in the browser on initial stage.
3. Click on the "Register" button to dynamically register Module B 100 times
4. Now we will see an increase in heap memory on account of registering module 100 times.
5. Click on "Un-Register" button to dynamically un-register Module B 100 times
6. Check the heap memory again, at this point will see an jump in memory usage on account of un-registering 100 module.

What is expected?

Heap memory generated on step 3 & 5 must be cleared after step 6. (Final stage)

What is actually happening?

Heap memory of the browser is not getting cleared.

---

We are observing memory leak of Store._vm while using registerModule/unregisterModule extensively.
On heap profiling we found that the oldVm instance is not getting garbage collected.

Probable Solution:
De-reference the oldVm instance on the same function scope where its getting instantiated (i.e. resetStoreVM() )

The above can be achieved by adding following code inside resetStoreVM function.

oldVm.$destroy();
oldVm = null; // This is the missing code

Attached google chrome (V 72.0.3626.96) heap snapshot and file.

Heap-20190213T220212.heaptimeline.zip

[parth67]

Closure function at https://github.com/vuejs/vuex/blob/dev/src/store.js#L259 is retaining environment which contains oldVm reference.

Check screenshot GC root is able to reach oldVm object

[tehnorm]

Possibly related. We are seeing a memory leak in a mutation:

[types.RECORD_MEASUREMENT](st, measurement) {
  Vue.set(st.lastMeasurementByDevice, measurement.device_id, measurement);
},

This seems to help the leak.

[types.RECORD_MEASUREMENT](st, measurement) {
  Vue.set(st.lastMeasurementByDevice, measurement.device_id, Object.freeze(measurement));
},

Still working on a succinct reproducible case.

[parth67]

@tehnorm can you share sample state declaration and sample measurement object?

[wolfgangwalther]

The fix in pr#1508 doesn't change anything for me, but the original suggestion by @skkart (adding oldVm = null) does solve the memory leak for me.

#1508 only fixed one of the anonymous functions preserving the previous context in resetStoreVM. See PR #1552 for a fix.

[afwn90cj93201nixr2e1re]

@ktsn what about the last one comment?

[wolfgangwalther]

finally resolved by #1546