WebServer memory leak (Rpi Pico W)

[brabl2]

Hi,
It looks there is a memory leak somewhere in the WebServer.
I only added this:
Serial.println(rp2040.getFreeHeap()); delay(20);
to the loop() in the AdvancedWebServer.ino example and found that the free heap decreases when any request is served by the WebServer.
According to my observations the free heap decreases by 88 bytes per each handled request regardless of data size sent out.

[revell1]

By adding a delay(20) you will also slow down the calls to "server.handleClient();".

Try removing that delay, and instead use a loop count to regulate your reporting of free heap.

[brabl2]

I replaced the reporting by this one:
if (++i>=100) {i=0; Serial.println(rp2040.getFreeHeap());}
However the behavior is same.
The free heap decreases by 88 bytes per each handled request.

[revell1]

OK.
Well that did not work, but atleast it will not be slowing the calls to handleClient().

Try replacing

if (++i>=100) {i=0; Serial.println(rp2040.getFreeHeap());}

With something like the following, where you only print the freeheap when i is 100, but print an integer value
when i is 200. Also using a variable to hold the heapfree value, rather than in the Serial.println call.
Should get the same rate of calls to println(), but half the rate of calls of getFreeHeap() so should see a change in the drift.

if (i==100) {numBytes = rp2040.getFreeHeap());Serial.println(numBytes);}
if (++i>=200) { i=0; Serial.println(12345);}

To see if it is the call to Serial.println() or rp2040.getFreeHeap() that is causing a problem.

I have been using rp2040.getFreeHeap() without issue to report heap space in a web server response, and it sits at a stable value (it initially reduces a little, but then is stable).

[brabl2]

It is a misunderstanding.
I'm not reporting issue in Serial.println() nor rp2040.getFreeHeap(). None of these two is causing the heap leak.
I'm reporting issue in WebServer.
I used the Serial.println(rp2040.getFreeHeap()); to show the leaking heap when using the AdvancedWebServer.ino example.
The size of free heap is approx. 190000 bytes at the start.
Based on my test each request handled by WebServer causes 88 bytes leak, so after approx. 2160 handled requests the AdvancedWebServer.ino example dies since there is no memory available.
I'm using the latest arduino-pico release 2.6.0

[revell1]

Hi, sorry, yes a bit of a misunderstanding, I had used this example previously and did not have a problem with memory leak, but I think that was back with version 2.5.2.

Now when I rebuild my SAVED project, it builds and runs, but if I add printing of heap, I also get an 88 byte reduction for each transfer request, i.e. the HTML and then the SVG file, so two times 88 per web page refresh.

So I can not tell if this is something that existed back with version 2.5.2 or has been added recently with some update to the code base, however I think this must be a new bug (see an old issue 860 that I raised some time back).

But I CAN CONFIRM your report.

@earlephilhower Earle, you probably recall issue 860 this memory leak could not have been present at that time as you were managing to run the EXAMPLE code for an extended period, and should have run out of memory during those tests. This would suggest the bug has been introduced since 2.5.2.

Also Earle, the code is using a variable "led" of 13, I guess this is a special GP pin output for an LED, would make sense if the onboard LED could be used for this indication, unless you are using the pin for some hardware scope test setup.

[earlephilhower]

Very odd, but reproducible on my end, too. I was using FireFox and I think that maybe it put the tab to sleep since it wasn't at the forefront so while the uptime was many hours on my prior testing, the actual number of page pulls wasn't so great.

[earlephilhower]

There was a lost client structure each connection.

I just laid on the F5 key and memory is very stable with the PR #914 (updated demo to include memory and page serve count)

[earlephilhower]

@brabl2 @revell1 I suggest you give #916 a try.

After I fixed the memory leak I was still getting random crashes or hangs, so I did a lot more debugging and think I have it pretty solid now with that PR.

[brabl2]

The WebServer reliability was improved with these two fixes (tested with Firefox and Vivaldi).
Thanks!