tobozo
commented
1 year ago hi, thanks for the feedback :+1:
Copying big files from LittleFS to PSRam could work if you don't create folders: the vfs implementation for folders is incomplete and may trigger bugs, but it should be fine if you only create files at the root of psramfs filesystem.
Another approach would be to maintain your own file struct and use RomDiskStream with ESPAsyncWebServer, preload your files at boot and even unmount LittleFs.
#include <LittleFS.h>
#include <PSRamFS.h>
#include <map>
void toPsram( fs::FS &sourceFS, const char* path );
struct PsRamFile_t
{
char* data;
size_t size;
RomDiskStream getStream() { return RomDiskStream((const uint8_t*)data, size); }
};
std::map<String,PsRamFile_t> MyPsramFiles; // global psram-files storage
void toPsram( fs::FS &sourceFS, const char* path )
{
fs::File myFile = sourceFS.open(path);
if (!myFile ) return;
auto data_mem = ((char*)ps_malloc(myFile.size()+1));
if( data_mem == NULL ) { // malloc failed !
myFile.close();
return;
}
PsRamFile_t myPsRamFile = { .data=data_mem, .size=myFile.size() };
size_t bytes_read = myFile.readBytes( myPsRamFile.data, myFile.size() );
if( myFile.size() != bytes_read ) {
// incomplete copy ?
Serial.printf("[WARNING] File copy missed %d bytes for %s\n", myFile.size()-bytes_read, path );
}
myFile.close();
MyPsramFiles[path] = myPsRamFile; // add file to global psram-files array
}
// somewhere in the ESPAsyncWebServer event loop
String myFilePath = "/path/to/myFile.txt";
if( MyPsramFiles[myFilePath] != NULL ) {
server.streamFile(MyPsramFiles[myFilePath].getStream(), "text/plain");
}
mhaberler
lbernstone
Hi,
this looks like a very interesting project.
I have a performance issue with a web server using ESPAsyncWebServer serving from a small LittleFS root (maybe 1MB) - which is so slow that reading large files triggers the watchdog no matter how long I set it
so I was thinking about caching the LittleFS content in a PSRAM-based filesystem
Do you think this project is a viable base for such an effort?
what I would do is:
another option would be to cook up something like a PSRAM-based LittleFS subclass which would just suck up the whole partition into PSRAM and serve from there, but I have no grasp how complex this would pan out.
Does this sound reasonable? which option would you choose?
I would be grateful for a frank opinion.
best regards Michael