[Ovmsdev] File logging task

[Michael Balzer]

Everyone,

I've pushed the long due refactoring of the file logging into a dedicated task. This removes most of the delays involved in file logging, as the fsync() and log
cycling now is done in the logging task and no longer blocks overall log message processing.

I stumbled upon a bad thing (tm) though we need to resolve with logging: the esp-idf logging generally involves a semaphore wait of max 10 ms (!). I wasn't
aware of this because I didn't look into the source before, there also is no warning about this in the esp-idf documentation
(https://docs.espressif.com/projects/esp-idf/en/latest/api-reference/system/log.html), so I guess I'm not the only one being taken by surprise here.

    // Maximum time to wait for the mutex in a logging statement.
    #define *MAX_MUTEX_WAIT_MS 10*
    #define MAX_MUTEX_WAIT_TICKS ((MAX_MUTEX_WAIT_MS + portTICK_PERIOD_MS - 1) / portTICK_PERIOD_MS)
    …
    void IRAM_ATTR *esp_log_write*(esp_log_level_t level,
            const char* tag,
            const char* format, ...)
    {
        if (!s_log_mutex) {
            s_log_mutex = xSemaphoreCreateMutex();
        }
        if (*xSemaphoreTake(s_log_mutex, MAX_MUTEX_WAIT_TICKS*) == pdFALSE) {
            return;
        }
    …


That semaphore is needed to secure the log level tag cache against parallel access, so it also affects log messages of currently disabled tags and/or levels.

This means we *must not* use the standard logging from timer callbacks, at no verbosity level. But we do, partially caused by my own code serving as a bad
example I assume -- sorry.

But I've not only found direct log calls in most vehicle timer callbacks, there also are some hidden ones from calling the framework, for example marking a
notification as read can produce a log output from Cleanup() if tracing is enabled.

We need to change this short to mid term. This may be the cause of those strange timer overflow issues we've seen occasionally. It also means every debug or
verbose log output currently in the code involves potentially multiple semaphore waits, that's for example the case for all hex dumps logged by the SIMCOM module.

Our primary option is to remove all log calls from all low level framework functions and timer callbacks.

A simple & quick workaround could be to extend the ESP_LOGx macros to check if they are executed in the timer task context, if so suppress the message or call
the ESP_EARLY log function instead. That will hide these log messages from the logging framework though, if using the _EARLY_ variant they only will be sent to
the USB port.

For a better solution I thought about adding a dedicated queue & task for the initial log submission as well. That will need quite some rework to the esp-idf
(no changes for this in the current master) but would allow to continue logging as we do now. Cons: queue submission involves some overhead & RAM, printf format
expansion needs to be done before the tag/level check, messages may need to be dropped if the task starves or RAM gets tight, log output to consoles will depend
on a task as well.

Opinions / better ideas?

Regards,
Michael

-- 
Michael Balzer * Helkenberger Weg 9 * D-58256 Ennepetal
Fon 02333 / 833 5735 * Handy 0176 / 206 989 26

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openvehicles.com/pipermail/ovmsdev/attachments/20191105/f1eaadfa/attachment.html>