feature request: Option to change the network timeout for Zephyr

[chirambaht]

One of the key function used when collecting data from the Memfault servers is prv_poll_socket. In this function, a constant, pre-defined value of 5000ms is used for the network connection timeout. In the ESP-IDF port, I saw that this has already been done for some of the other networking functions using CONFIG_MEMFAULT_HTTP_CLIENT_TIMEOUT_MS. It would be nice to also have this value be defined by a Kconfig in much the same way for the Zephyr port as well.

[chirambaht]

To add more context to this issue:
I have a device which I am currently trying to update and quite often, the device reports back the following logs:

[00:14:07.288,000] <dbg> mflt: memfault_platform_log: Downloaded 62976/2097152 bytes (3%)
[00:14:16.726,000] <err> mflt: Timeout waiting for socket event(s): event(s)=1, errno=11
[00:14:16.730,000] <err> mflt: Error upgrading firmware, rv=-1
[00:12:49.693,000] <err> net_tcp: conn: 0x3fcb7278, new bytes 2825 during TIME-WAIT state sending reset

I then decided to increase the timeout manually and found that my device could now download the image on its first try. I was simultaneously running a device which took multiple attempts to download the same image (same cohort). The higher timeout downloaded it on its first try while the device with the default timeout eventually downloaded the payload but after multiple re-attempts.

[noahp]

@chirambaht thanks for reporting! great idea, we will add it to the next SDK release. Let me know if there's urgency for a new SDK with the changes, and we can expedite it- otherwise, it's scheduled for sometime next week.

[noahp]

Hi @chirambaht - just to confirm, is this change what you were thinking of?

diff --git a/ports/zephyr/Kconfig b/ports/zephyr/Kconfig
index db6b65bd7b..c9c75eb2c6 100644
--- a/ports/zephyr/Kconfig
+++ b/ports/zephyr/Kconfig
@@ -403,6 +403,14 @@ config MEMFAULT_PROJECT_KEY
           Token used to communicate with Memfault. Find it at
           https://mflt.io/project-key
 
+config MEMFAULT_HTTP_CLIENT_TIMEOUT_MS
+        int "The HTTP client timeout in milliseconds"
+        default 5000
+        help
+          The Memfault HTTP client timeout in milliseconds. This is the
+          maximum amount of time the HTTP client will wait for a response from
+          the server.
+
 endif # MEMFAULT_HTTP_ENABLE
 
 config MEMFAULT_ZEPHYR_FOTA_DOWNLOAD_CALLBACK_CUSTOM
diff --git a/ports/zephyr/common/memfault_platform_http.c b/ports/zephyr/common/memfault_platform_http.c
index c368181af3..8b7bae9155 100644
--- a/ports/zephyr/common/memfault_platform_http.c
+++ b/ports/zephyr/common/memfault_platform_http.c
@@ -75,7 +75,7 @@
 #endif /* CONFIG_MEMFAULT_HTTP_USES_MBEDTLS */
 
 // Timeout for the socket file descriptor to become available for I/0
-#define POLL_TIMEOUT_MS 5000
+#define POLL_TIMEOUT_MS CONFIG_MEMFAULT_HTTP_CLIENT_TIMEOUT_MS
 
 // The nRF-Connect SDK has an implementation of this structure
 #if !defined(CONFIG_MEMFAULT_NCS_PROJECT_KEY)

[noahp]

This is fixed in 1.20.0. Let me know if there's any issues!