fds_gc() does not skip pages after first usage of FDS

[Elias]

I'm trying to debug issues in my application related to FDS garbage collection.

My application periodically runs fds_gc() from startup. At this point, garbage collection is doing nothing. However, after I call pm_peer_rank_highest() once, fds_gc() appears to garbage-collect the page each time it is run. Adding some debug output to nrf_fstorage_erase(), I see that it is called after every fds_gc() after pm_peer_rank_highest(), and never before. Digging into fds.c, I see where m_pages[].can_gc is set to 'true', but I don't see where it is set to 'false'. Is this a bug?

I'm using nRF SDK 14.2, with Peer Manager and FDS modules backported from SDK 16.

[Einar Thorsrud]

Hi,

Thank you for reporting. It seems you are right that this is a bug. Referring to FDS from SDK 16, m_pages[page].can_gc is set to true on line 830 and 958 in fds.c. It is is also set on line 722, by the call to page_scan() if a dirty record exists, but it is never set to false. I have reported it to the SDK team.

I do not see this as a major bug though, and it should not have any significant consequences. 

[Elias]

The consequence I was worried about was flash wear. I originally planned to have fds_gc() run once per minute, but this will go through the 10,000 write-erase cycles for flash in just under a week. Will this thread be updated once a fix is developed?

[Einar Thorsrud]

Hi,

Elias said:

The consequence I was worried about was flash wear. I originally planned to have fds_gc() run once per minute, but this will go through the 10,000 write-erase cycles for flash in just under a week

To be honest I did not think of running GC that often. Why do you want to do that? Typically we recomend do run it when you get FDS_ERR_NO_SPACE_IN_FLASH returned when writing of updating a record (also if you get NRF_ERROR_STORAGE_FULL returned from certain peer manager calls, which uses FDS). This minimizes the number of garbage collection runs, saving power and reducing flash wear.

Update: If you do not want to wait for the FDS to be full before doing GC, you can check the amount of free space using fds_stat() regularly, and make a desition on wheater to do GC based on that.

Elias said:

Will this thread be updated once a fix is developed?

I cannot promise that this will be part of the next nRF5 SDK release, so it might take some time. I will try to remember updating this thread if/when it happens, though.

[Elias]

I prefer to prevent non-NRF_SUCCESS return codes whenever possible. The fds_stat() strategy sounds like it will work.