Unexpected behaviour of FileInterface::getContents method in reactphp/filesystem

[odmyikona]

Hi there,

Thanks once again for this wonderful software.

Is it the expected behaviour of FileInterface::getContents method not to reflect changes if it is called once or am I missing something?

Under this series of events the result of the final step is different from what's expected:

• Running a loop
• Get file content with FileInterface::getContents
• Append to file with FileInterface::putContents (using await)
• Get file contents again with FileInterface::getContents, but the content do not reflect the changes made in previous step

Note: file_get_contents works.

I am running "Framework-X" in a docker container as such:

FROM php:8.1-alpine3.16

RUN apk --no-cache add ${PHPIZE_DEPS} libev \
    && pecl install ev \
    && docker-php-ext-enable ev \
    && docker-php-ext-install sockets \
    && docker-php-ext-install pcntl \
    && echo "memory_limit = -1" >> "${PHP_INI_DIR}/conf.d/acme.ini"

[...]

Thanks!

[WyriHaximus]

Hey @odmyikona,

Glad to hear your enjoy using our project :D

Going to assume you're using the 0.2 branch. The 0.2 branch is doing a stat before a read, currently working on a PR that will remove that and test a few situations that might occur while reading from a file. The issue could be that that stat is cached and as such still thinks the file is the old size.

[kristos80]

Hi @WyriHaximus thanks for the answer. Mistakenly I have started the discussion from our company's robot account.

Yes, I can confirm that I am using the ^0.2@dev branch, but this unexpected behaviour is seen only when using the FileInterface with "Framework-X" and not when used alone.

I'll come back with more in the first chance.

[odmyikona]

@WyriHaximus yes if patch the file directly with your code, it does work as expected!

[kristos80]

@WyriHaximus thanks also for the image suggestion as well. Actually I have tried to create this current image of mine by "stealing" some ideas from yours several months ago, but it seems that I didn't do a good job :)

[WyriHaximus]

@WyriHaximus yes if patch the file directly with your code, it does work as expected!

Great! My theory is that after a few times that stat call is cached so it doesn't check anymore for a while. After a while you should start seeing the rest of the file again once the cache expires. Just a theory, tho.

@WyriHaximus thanks also for the image suggestion as well. Actually I have tried to create this current image of mine by "stealing" some ideas from yours several months ago, but it seems that I didn't do a good job :)

Feel free to post feedback on why you didn't go with my images on my repo for it. Not sure if this thread is the best place to do that. Assuming you have any :).

[kristos80]

@WyriHaximus even after 3 hours it didn't had this sort of caching busted. I was fetching that file from within the same loop, but always the same cached result. If I restart the loop then I get the changes, but gets immediately cached as well. The only way for emptying this sort of cache is to delete/recreate the file outside of the loop, where again it would be cached after some tries to read it. Really intriguing as you said since this is the fallback mechanism.

I really am not sure why I didn't go with the exact image (probably for teaching myself on how to do that alone), but I was confident that I had ext-uv installed in the current image. This is a big lesson to learn and try not to re-invent the wheel (sorry for this off-topic)

[WyriHaximus]

@WyriHaximus even after 3 hours it didn't had this sort of caching busted. I was fetching that file from within the same loop, but always the same cached result. If I restart the loop then I get the changes, but gets immediately cached as well. The only way for emptying this sort of cache is to delete/recreate the file outside of the loop, where again it would be cached after some tries to read it.

@kristos80 Good to know I was wrong with that theory.

Really intriguing as you said since this is the fallback mechanism.

Yes, but because the stat gets cache it keeps reporting the old file size. In theory you should observe the same behavior when using ext-uv or ext-eio. But working on fixing that and making sure it becomes more reliable :).

I really am not sure why I didn't go with the exact image (probably for teaching myself on how to do that alone), but I was confident that I had ext-uv installed in the current image. This is a big lesson to learn and try not to re-invent the wheel (sorry for this off-topic)

Learning is a great reason to do things. That's how I ended up in this project ;).