Is there any way to write from one client and read from other client via blobfuse2?

[jejass]

Hello team,

Here is the scenario. Both client A and client B mount Storage account via blobfuse2. Client A is responsible for write, Client B is responsible for read.

I test 2 configurations, with both of them, client B cannot read the latest updated blob, even after waiting for 1 hour.

configuration 1:

logging:
type: syslog
level: log_debug

components:

• libfuse
• file_cache
• attr_cache
• azstorage

libfuse:
attribute-expiration-sec: 0
entry-expiration-sec: 0
negative-entry-expiration-sec: 0

file_cache:
path: /mnt/blobfusetmp
timeout-sec: 0
max-size-mb: 0
refresh-sec: 1

attr_cache:
timeout-sec: 0

configuration 2:
logging:
type: syslog
level: log_debug

components:

• libfuse
• file_cache
• attr_cache
• azstorage

libfuse:
attribute-expiration-sec: 0
entry-expiration-sec: 0
negative-entry-expiration-sec: 0

block_cache:
block-size-mb: 1
mem-size-mb: 1000
path: /mnt/blobfusetmp
disk-size-mb: 1
disk-timeout-sec: 1

attr_cache:
timeout-sec: 0

I understand 'direct_io' parameter can disable kernal page cache. but with this parameter, the application will report an error.

So, I want to figure out a way in the blobfuse2 configuration to make Client B can read the latest update.

is that possible?

as per my testing, it seems the Client B will read the blob once. After then, the read Client B will not refresh the content, even with timeout 0, cache size 0. I found the parameter 'refresh-sec': . but it does not work as well.

is that how blobfuse2 works? the read client will never refresh the content in any configuration?

thank you in advance.

[souravgupta-msft]

Hi @jejass. Thanks for reaching out. You are not seeing updated data in Client B because of kernel caching. To disable kernel cache, you will have to use direct_io or manually clear the kernel cache using the below command before read operations,
echo 2 > /proc/sys/vm/drop_caches

[jejass]

@souravgupta-msft ,thanks for your response. the thing is here, cx is not able to use direct_io as their application will report error message when using direct_io. It looks like something wrong due to the C++ shared memory model.

is there any way from blobfuse side to make the kernal cache expiry then reload the data from Storage account?

[souravgupta-msft]

Can you try with this parameter, --disable-writeback-cache=true