Remove .with_current_subscriber() calls
#992
Conversation
|
|
I'm not entirely sure. I might have discussed the original problem long ago with Gor and/or @wprzytula. I think it might have been necessary for the cpp-rust-wrapper. One use case that comes to my mind is:
|
|
Might be related to this: scylladb/cpp-rust-driver#110 |
Indeed, this cpp-rust-driver's logging might malfunction after those calls are removed. @Lorak-mmk please build the cpp-rust-driver against custom rust driver with those calls removed and check that logging still works. |
|
If I understand correctly this is not in issue when global default subscriber is used, because it is a default for all threads. This is what cpp-rust-driver currently does. |
wprzytula
removed
the
waiting-on-author
There was a problem hiding this comment.
Cargo.lock.msrv needs to be updated. Otherwise, looks acceptable to me, although I'm still reluctant to believe that cpp-rust-driver's logging really works without those calls; they really had some purpose and I feel that there is some trap that we forget about.
If there is something that I didn't notice I'd prefer to work around it in cpp-rust-driver (even if it's inconvenient or even requires a patched version of the driver) rather than break the experience for Rust Driver users. |
Lorak-mmk
force-pushed
the
remove_with_current_subscriber
branch
from
20c8002 to
dd457c7
Compare
Those calls were introduced in e557095, with message: ``` Now, when the driver spawns a task to run a new future on it, that future will use the same subscriber as the code that spawned the task in the first place. ``` There is unfortunately no explanation about when it is necessary. I don't see any problems after removing those - logs are still collected correctly using a tracing subscriber. Those calls however have a harmful side-effect: they prevent usage of `log` loggers to listen to driver logs using `log` feature in `tracing` crate. After reporting the problem to `tracing` crate: tokio-rs/tracing#2913 one of maintainers said that using `.with_current_subscriber()` in a library is not necessary in general. As I don't see any issue caused by removing these calls, but their existence cause an issue, they are removed in this commit.
This example demonstrates usage of `env_logger` to listen to driver logs.
Lorak-mmk
force-pushed
the
remove_with_current_subscriber
branch
from
dd457c7 to
2199bda
Compare
Lorak-mmk
force-pushed
the
remove_with_current_subscriber
branch
from
2199bda to
370b00a
Compare
| You can use `cargo add tracing -F log` or edit `Cargo.toml`: | ||
| ```toml | ||
| tracing = { version = "0.1.40" , features = ["log"] } | ||
| ``` | ||
| then you can setup `env_logger` os some other logger and it will output logs from the driver: |
There was a problem hiding this comment.
Will enabling log feature in the end crate's Cargo.toml suffice, or is it required to have it enabled for scylla crate as well?
Another question. What if the end user crate uses tracing crate in version that is incompatible with the version that scylla crate uses?
There was a problem hiding this comment.
- Yes, it is enough
- Not sure what you are asking. Are you asking if adding hypotetical tracing 0.2 to end user Cargo.toml and enabling log feature there will enable log feature in our tracing 0.1? No, it won't
There was a problem hiding this comment.
Quoting comment from the tracing issue I opened (tokio-rs/tracing#2913 (comment) ):
A library should not generally enable the log or log-always feature flags, and generally should never be responsible for setting the default subscriber. Instead, user code that depends on the library should enable tracing's log crate support if it wants to, and set the default subscriber if it wishes to use tracing. I don't think the calls to with_current_subscriber are necessary in a library in general.
There was a problem hiding this comment.
- I'm asking about whether mismatched versions of
tracinginscyllacrate and the end crate can cause incompatility issues, resulting in misbehaving tracing.
There was a problem hiding this comment.
No idea, probably. It's a question to tracing crate maintainers, how are they going to handle new major version to avoid ecosystem split.
…bscriber Remove `.with_current_subscriber()` calls (cherry picked from commit 88bc8c1)
…bscriber Remove `.with_current_subscriber()` calls (cherry picked from commit 88bc8c1)
Those calls were introduced in e557095,
with message:
There is unfortunately no explanation about when it is necessary.
I don't see any problems after removing those - logs are still collected correctly using a tracing subscriber.
Those calls however have a harmful side-effect: they prevent usage of
logloggers to listen to driver logs usinglogfeature intracingcrate, which was reported in #860 .After reporting the problem to
tracingcrate (tokio-rs/tracing#2913) one of maintainers said that using.with_current_subscriber()in a library is not necessary in general.As I don't see any issue caused by removing these calls, but their existence cause an issue, they are removed in this PR.
The PR also introduces an example showing usage of the driver with
env_loggerand adjusts documentation about logging.Fixes: #860
@piodul if you remember why those calls were added please explain, because I have no idea in which scenario they could be necessary.
Pre-review checklist
./docs/source/.Fixes:annotations to PR description.