Data loss if issues on TCP protocol layer or failures on network link. Retry policy is ignored

[YuriyHolinko]

Is your feature request related to a problem? Please describe.
I have noticed data loss in shipping of logs even if retry policy is used. the problem is that we do not have any strong logic around retryable exceptions and if there is some issue in network link, the data is not re-sent instead, the exception is silently swallowed.
the code which decides retry or not is here

opentelemetry-java/exporters/sender/okhttp/src/main/java/io/opentelemetry/exporter/sender/okhttp/internal/RetryInterceptor.java

Line 148 in 1fcbed8

static boolean isRetryableException(IOException e) {

however, it does not handle all the issues with network link failures e.g. my use case

java.net.SocketTimeoutException: timeout
because the message does not match the condition
message == null || message.toLowerCase(Locale.ROOT).contains("connect timed out");

Describe the solution you'd like
I suggest modifying the condition

• to cover more cases
• to make it configurable at the programmatic level, so users can adapt it to the specific network failures they may encounter

Describe alternatives you've considered
I cannot define

Additional context
I can bring a draft PR with my fix as I am not using OTEL java agents to ship logs but initiate it programmatically . hope this will be useful for others as well

[jack-berg]

Thanks for the report.

I recommend for now we adjust the logic to catch this case you've found. If it becomes a recurring problem, we can consider making it programmatically configurable, but let's opt for the simple solution for now.

PRs welcome!

[YuriyHolinko]

thanks for reply

in my case it will be 3 exceptions when I want to retry. if we use SSL under the hood there will be more, because when servers (log receivers) are crashed/failed, ssl handshake exceptions might be thrown and we need to handle it as well. that's why I want to make it configurable

in any case I start following procedure how to become a contributor and get a member access to repo to create my first PR

[trask]

how to become a contributor and get a member access to repo to create my first PR

hi @YuriyHolinko! no special access needed, we recommend to fork the repo and send PR from your fork

[jack-berg]

Keeping open despite merging #6991 because I think we should update the default retry exception predicate to include some additional exceptions, as discussed here.

[YuriyHolinko]

covered by
#6991
#7047
#7057