UDP echotests fix in case of no memory or device busy.

[tymoteuszblochmobica]

UDP echotests hold in case of memory or device busy status.
This gives possibility of freeing memory and mesh device recover from busy state.

Summary of changes

If Wisun mesh is running and mbed-trace is disabled

UDP burst tests consumes all available memory and sendto fails with NSAPI_ERROR_NO_MEMORY because Nanostack ns_mem_internal_alloc returns NULL buffer.
Problem is with temporary memory lack this fix gives time to memory manager for freeing unused memory pools.

Impact of changes

No

Migration actions required

No

Documentation

---

Pull request type

[x] Patch update (Bug fix / Target update / Docs update / Test update / Refactor)
[] Feature update (New feature / Functionality change / New API)
[] Major update (Breaking change E.g. Return code change / API behaviour change)

---

Test results

[] No Tests required for this change (E.g docs only update)
[x] Covered by existing mbed-os tests (Greentea or Unittest)
[] Tests / results supplied as part of this PR

---

Reviewers

@SeppoTakalo
@AnttiKauppila
@michalpasztamobica

[michalpasztamobica]

Don't we need to do the same for the non-blocking implementation?
And don't we need to check for NSAPI_ERROR_BUSY here?

[SeppoTakalo]

Lets re-think this..

Your sleep_for() might be fixing the issue, but only because it adds some delay.

But within that delay, there might be other packets coming (duplicates), cumulating our input buffers.

So instead of adding a error handling here, we should follow a generic UDP application priciples:

1. Ignore send errors.
2. Don't retry immediately.
3. Let upper layer or application logic handle the retries and timeouts.

Our application logic is a few lines down. On line 103 on original. There is continue call, which immediately retries. So to fix this, I propose that we remove the continue and allow it to flow directly into the receive() loop that starts from line 106.

Notice that in case the packet did not really go out, we get nothing back, and this loop will restart. So the retry logic is there already. But instead of immediately retrying, it would allow recv() or recvfrom() to drain potential duplicate packets away from input buffers.

So, we should be OK with only one "sleep" within this logic. If this does not work, make sure that SOCKET_TIMEOUT is long enough that we do sleep in recvfrom() and while doing that, drain our input buffers.

[tymoteuszblochmobica]

Done as you suggested. Both continue and sleep removed.
It's also working fine.

[ciarmcom]

@tymoteuszblochmobica, thank you for your changes.
@AnttiKauppila @SeppoTakalo @michalpasztamobica @ARMmbed/mbed-os-test @ARMmbed/mbed-os-ipcore @ARMmbed/mbed-os-maintainers please review.

[SeppoTakalo]

Lets re-think this..

Your sleep_for() might be fixing the issue, but only because it adds some delay.

But within that delay, there might be other packets coming (duplicates), cumulating our input buffers.

So instead of adding a error handling here, we should follow a generic UDP application priciples:

1. Ignore send errors.
2. Don't retry immediately.
3. Let upper layer or application logic handle the retries and timeouts.

Our application logic is a few lines down. On line 103 on original. There is continue call, which immediately retries. So to fix this, I propose that we remove the continue and allow it to flow directly into the receive() loop that starts from line 106.

Notice that in case the packet did not really go out, we get nothing back, and this loop will restart. So the retry logic is there already. But instead of immediately retrying, it would allow recv() or recvfrom() to drain potential duplicate packets away from input buffers.

So, we should be OK with only one "sleep" within this logic. If this does not work, make sure that SOCKET_TIMEOUT is long enough that we do sleep in recvfrom() and while doing that, drain our input buffers.

Pull request has been modified.

[tymoteuszblochmobica]

Done as Seppo suggested. Both continue and sleep removed.
It's also working fine.

[jamesbeyond]

CI started

[mbed-ci]

Test run: SUCCESS

Summary: 5 of 5 test jobs passed
Build number : 1
Build artifacts

[mergify]

This PR does not contain release version label after merging.

[0xc0170]

Investigating why the label is being not picked up :/