Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Api in WiFiClient class for saving heap memory #8738

Merged
merged 3 commits into from
Dec 6, 2022

Conversation

RobertGnz
Copy link
Contributor

@RobertGnz RobertGnz commented Dec 2, 2022

Api for saving heap when Client class is used by a Server (WiFiServer class): Client = Server.available().

Suppose the local end is the server and the remote end is the client, we will deal with heap memory at the local end.

When the local application (server) decides to close an active connection with a remote end it issues an Client.stop.
The stop() function calls the close() function of ClientContext class which in turn calls tcp_close.
The connexion is closed by tcp_close and the protocol control block (pcb) can be put in the following states depending on the requests sent by the remote: CLOSING, FIN_WAIT_1 and FIN_WAIT_2. In theses states pcbs are not freed, then consume some memory heap.
If an acknowledgment from the remote end is received, the pcb enter in TIME_WAIT state for some minutes but pcbs in TIME_WAIT state are not freed. Then consume some heap memory.
TIME_WAIT pcbs are automatically freed after some minutes or can be freed for instance issuing an tcp_kill_timewait()
in the local application which will free the oldest pcb in TIME_WAIT state.

If the connection is first closed from the remote end (the client), the local end (server) receive a connection termination request. It then acknowledge it and enter in CLOSE_WAIT state waiting for a connection termination request from the local application.
It then send a termination request and enter in LAST_ACK state until it receive an acknowledgment from the remote end.
After receiving the acknowledgment it enter in ClOSED state and the local pcb is freed leaving some room in the heap memory.

To summarize, when a connexion termination request is send by one end (remote or local), the local pcb is not freed immediatly.
This pcb can be in the following states: FIN_WAIT_1, FIN_WAIT_2, CLOSING, TIME_WAIT, CLOSE_WAIT, LAST_ACK.
As a consequence, some old pcbs from old closed connections are still consuming heap memory.

The local application can call tcp_kill_timewait hoping it will free some TIME_WAIT state pcbs. But if the server receive frequent connections requests and close them after sending whatever it has to send, there may be zero pcbs in TIME_WAIT state among all previously closed connections.

In case of insufficient memory to accept a new connection, lwip has developped a strategy: it successively tries to kill the oldest pcb in TIME_WAIT state, or in LAST_ACK state or in CLOSING state or the oldest active connection with lower priority than the new one.

As a matter of fact this "urgent" strategy is deployed only when very few heap memory remain available (less than some kb). In case of success, Client.available returns a valid Client but the local application will crash when sending or receiving data from the client (Client.read ou readuntil or available) because this need more heap memory and just some kb were freed in lwip to allocate the new pcb structure ans start the new connection.

The propose API is intended to avoid this drawback by calling the abort function of ClientContext which in turn calls tcp_abort which calls tcp_abandon. The connection is aborted and notified to the client with a RESET and the pcb and ressources associated are immediately released increasing the available heap memory.

This API can be used in two ways:

WiFiClient Client;

1- Replace every Client.stop() with Client.abort()
or
2- In conjonction with Client.stop in the following style:
# define MIN_HEAP_FREE 20000 // or whatever min available heap memory convienent for your application
if ( ESP.getFreeHeap() >= MIN_HEAP_FREE ) Client.stop();
else Client.abort();

…nd acting as server

The propose API is intended to avoid this drawback by calling the abort function of ClientContext which in turn calls tcp_abort which calls tcp_abandon. The connection is aborted and notified to the client with a RESET and the pcb and ressources associated are immediately released increasing the available heap memory.
@mcspr
Copy link
Collaborator

mcspr commented Dec 3, 2022

Thanks!
I would still move comment / api description to the web documentation (https://github.com/esp8266/Arduino/blob/master/doc/esp8266wifi/client-class.rst) instead, unlike lwip we do not autogenerate things. Will edit text a bit, then push to our tree

PR still has a full text, API docs preferably should not contain too
much of impl details that may or may not be relevant to a common user
@mcspr mcspr merged commit 3c6db4e into esp8266:master Dec 6, 2022
hasenradball pushed a commit to hasenradball/Arduino that referenced this pull request Nov 18, 2024
Api for saving heap when Client class is used by a Server (WiFiServer class): Client = Server.available().

Suppose the local end is the server and the remote end is the client, we will deal with heap memory at the local end.

When the local application (server) decides to close an active connection with a remote end it issues an Client.stop.
The stop() function calls the close() function of ClientContext class which in turn calls tcp_close.
The connexion is closed by tcp_close and the protocol control block (pcb) can be put in the following states depending on the requests sent by the remote: CLOSING, FIN_WAIT_1 and FIN_WAIT_2. In theses states pcbs are not freed, then consume some memory heap.
If an acknowledgment from the remote end is received, the pcb enter in TIME_WAIT state for some minutes but pcbs in TIME_WAIT state are not freed. Then consume some heap memory.
TIME_WAIT pcbs are automatically freed after some minutes or can be freed for instance issuing an tcp_kill_timewait()
in the local application which will free the oldest pcb in TIME_WAIT state.

If the connection is first closed from the remote end (the client), the local end (server) receive a connection termination request. It then acknowledge it and enter in CLOSE_WAIT state waiting for a connection termination request from the local application.
It then send a termination request and enter in LAST_ACK state until it receive an acknowledgment from the remote end.
After receiving the acknowledgment it enter in ClOSED state and the local pcb is freed leaving some room in the heap memory.

To summarize, when a connexion termination request is send by one end (remote or local), the local pcb is not freed immediatly.
This pcb can be in the following states: FIN_WAIT_1, FIN_WAIT_2, CLOSING, TIME_WAIT, CLOSE_WAIT, LAST_ACK.
As a consequence, some old pcbs from old closed connections are still consuming heap memory.

The local application can call tcp_kill_timewait hoping it will free some TIME_WAIT state pcbs. But if the server receive frequent connections requests and close them after sending whatever it has to send, there may be zero pcbs in TIME_WAIT state among all previously closed connections.

In case of insufficient memory to accept a new connection, lwip has developped a strategy: it successively tries to kill the oldest pcb in TIME_WAIT state, or in LAST_ACK state or in CLOSING state or the oldest active connection with lower priority than the new one.

As a matter of fact this "urgent" strategy is deployed only when very few heap memory remain available (less than some kb). In case of success, Client.available returns a valid Client but the local application will crash when sending or receiving data from the client (Client.read ou readuntil or available) because this need more heap memory and just some kb were freed in lwip to allocate the new pcb structure ans start the new connection.

The propose API is intended to avoid this drawback by calling the abort function of ClientContext which in turn calls tcp_abort which calls tcp_abandon. The connection is aborted and notified to the client with a RESET and the pcb and ressources associated are immediately released increasing the available heap memory.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants