Crates.io | diesel-tracing |
lib.rs | diesel-tracing |
version | 0.3.1 |
source | src |
created_at | 2020-08-12 15:20:07.230317 |
updated_at | 2024-11-11 15:04:44.857052 |
description | Connection telemetry middleware for diesel and tracing |
homepage | |
repository | https://github.com/CQCL/diesel-tracing |
max_upload_size | |
id | 275839 |
size | 40,566 |
diesel-tracing
provides connection structures that can be used as drop in
replacements for diesel connections with extra tracing and logging.
Just like diesel this crate relies on some feature flags to specify which
database driver to support. Just as in diesel configure this in your
Cargo.toml
[dependencies]
diesel-tracing = { version = "<version>", features = ["<postgres|mysql|sqlite>"] }
diesel-tracing
has several instrumented connection structs that wrap the underlying
diesel
implementations of the connection. As these structs also implement the
diesel::Connection
trait, establishing a connection is done in the same way as
the diesel
crate. For example, with the postgres
feature flag:
#[cfg(feature = "postgres")]
{
use diesel_tracing::pg::InstrumentedPgConnection;
let conn = InstrumentedPgConnection::establish("postgresql://example");
}
This connection can then be used with diesel dsl methods such as
diesel::prelude::RunQueryDsl::execute
or diesel::prelude::RunQueryDsl::get_results
.
In some applications it may be desirable to be able to use both instrumented and
uninstrumented connections. For example, in the tests for a library. To achieve this
you can use the diesel::Connection
trait.
fn use_connection(
conn: &impl diesel::Connection<Backend = diesel::pg::Pg>,
) -> () {}
Will accept both diesel::PgConnection
and the InstrumentedPgConnection
provided by this crate and this works similarly for other implementations
of Connection
if you change the parametized Backend marker in the
function signature.
Unfortunately there are some methods specific to backends which are not
encapsulated by the diesel::Connection
trait, so in those places it is
likely that you will just need to replace your connection type with the
Instrumented version.
diesel-tracing
supports the r2d2
connection pool, through the r2d2
feature flag. See diesel::r2d2
for details of usage.
Currently the few fields that are recorded are a subset of the OpenTelemetry
semantic conventions for databases.
This was chosen for compatibility with the tracing-opentelemetry
crate, but
if it makes sense for other standards to be available this could be set by
feature flag later.
Database statements may optionally be recorded by enabling the
statement-fields
feature. This uses diesel::debug_query
to convert the query into a string. As this may expose sensitive information,
the feature is not enabled by default.
It would be quite useful to be able to parse connection strings to be able to provide more information, but this may be difficult if it requires use of diesel feature flags by default to access the underlying C bindings.
All logged traces are currently set to DEBUG level, potentially this could be changed to a different default or set to be configured by feature flags. At them moment this crate is quite new and it's unclear what a sensible default would be.
Errors in Result objects returned by methods on the connection should be
automatically logged through the err
directive in the instrument
macro.
As statements may contain sensitive information they are currently not recorded
explicitly, unless you opt in by enabling the statement-fields
feature.
Finding a way to filter statements intelligently to solve this problem is a
TODO.
Similarly connection strings are not recorded in spans as they may contain passwords
Record and log connection information (filtering out sensitive fields)
Provide a way of filtering statements, maybe based on regex?
License: MIT