-
Notifications
You must be signed in to change notification settings - Fork 1.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
feat(codecs): Add syslog encoder #21307
Draft
polarathene
wants to merge
16
commits into
vectordotdev:master
Choose a base branch
from
polarathene:feat/syslog-codec
base: master
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
Draft
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original commit from syedriko
This is only a temporary change to make the diffs for future commits easier to follow.
- Introduce a `Pri` struct with fields for severity and facility as enum values. - `Pri` uses `strum` crate to parse string values into their appropriate enum variant. - Handles the responsibility of encoding the two enum values ordinal values into the `PRIVAL` value for the encoder. - As `Facility` and `Severity` enums better represent their ordinal mapping directly - The `Fixed` + `Field` subtyping with custom deserializer isn't necessary. Parsing a string that represents the enum by name or its ordinal representation is much simpler. - Likewise this removes the need for the get methods as the enum can provide both the `String` or `u8` representation as needed.
`SyslogSerializer::encode()` has been simplified. - Only matching `Event::Log` is relevant, an `if let` bind instead of `match` helps remove a redundant level of nesting. - This method only focuses on boilerplate now, delegating the rest to `ConfigDecanter` (_adapt `LogEvent` + encoder config_) and `SyslogMessage` (_encode into syslog message string_). - This removes some complexity during actual encoding logic, which should only be concerned about directly encoding from one representation to another, not complimentary features related to Vector config or it's type system. The new `ConfigDecanter` is where many of the original helper methods that were used by `SyslogSerializer::encode()` now reside. This change better communicates the scope of their usage. - Any interaction with `LogEvent` is now contained within the methods of this new struct. Likewise for the consumption of the encoder configuration (instead of queries to config throughout encoding). - The `decant_config()` method better illustrates an overview of the data we're encoding and where that's being sourced from via the new `SyslogMessage` struct, which splits off the actual encoding responsibility (see next commit).
`SyslogSerializerConfig` has been simplified. - Facility / Severity deserializer methods aren't needed, as per their prior refactor with `strum`. - The `app_name` default is set via `decant_config()` when not configured explicitly. - The other two fields calling a `default_nil_value()` method instead use an option value which encodes `None` into the expected `-` value. - Everything else does not need a serde attribute to apply a default, the `Default` trait on the struct is sufficient. - `trim_prefix` was removed as it didn't seem relevant. `tag` was also removed as it's represented by several subfields in RFC 5424 which RFC 3164 can also use. `SyslogMessage::encode()` refactors the original PR encoding logic: - Syslog Header fields focused, the PRI and final message value have already been prepared prior. They are only referenced at the end of `encode()` to combine into the final string output. - While less efficient than `push_str()`, each match variant has a clear structure returned via the array `join(" ")` which minimizes the noise of `SP` from the original PR. Value preparation prior to this is clear and better documented. - `Tag` is a child struct to keep the main logic easy to grok. `StructuredData` is a similar case.
No changes beyond relocating the code into a single file.
- Drop notes referring to original PR differences + StructuredData adaption references. None of it should be relevant going forward. - Revise some other notes. - Drop `add_log_source` method (introduced from the original PR author) in favor of using `StructuredData` support instead.
This should be simple and lightweight enough to justify for the DRY benefit? This way the method doesn't need to be duplicated redundantly. That was required because there is no trait for `FromRepr` provided via `strum`. That would require a similar amount of lines for the small duplication here. The `akin` macro duplicates the `impl` block for each value in the `&enums` array.
- `ConfigDecanter::get_message()` replaces the fallback method in favor of `to_string_lossy()` (a dedicated equivalent for converting `Value` type to a String type (_technically it is a CoW str, hence the follow-up with `to_string()`_)). - This also encodes the value better, especially for the default `log_namespace: false` as the message value (when `String`) is not quote wrapped, which matches the behaviour of the `text` encoder output. - Additionally uses the `LogEvent` method `get_message()` directly from `lib/vector-core/src/event /log_event.rs`. This can better retrieve the log message regardless of the `log_namespace` setting. - Encoding of RFC 5424 fields has changed to inline the `version` constant directly, instead of via a redundant variable. If there's ever multiple versions that need to be supported, it could be addressed then. - The RFC 5424 timestamp has a max precision of microseconds, thus this should be rounded and `AutoSi` can be used (_or `Micros` if it should have fixed padding instead of truncating trailing `000`_).
- The original PR author appears to have relied on a hard-coded timestamp key here. - `DateTime<Local>` would render the timestamp field with the local timezone offset, but other than that `DateTime<Utc>` would seem more consistent with usage in Vector, especially since any original TZ context is lost by this point? - Notes adjusted accordingly, with added TODO query for each encoding mode to potentially support configurable timezone.
- Move encoder config settings under a single `syslog` config field. This better mirrors configuration options for existing encoders like Avro and CSV. - `ConfigDecanter::value_by_key()` appears to accomplish roughly the same as the existing helper method `to_string_lossy()`. Prefer that instead. This also makes the `StructuredData` helper `value_to_string()` redundant too at a glance? - Added some reference for the priority value `PRIVAL`. - `Pri::from_str_variants()` uses the existing defaults for fallback, communicate that more clearly. Contextual note is no longer useful, removed.
To better communicate the allowed values, these two config fields can change from the `String` type to their appropriate enum type. - This relies on serde to deserialize the config value to the enum which adds a bit more noise to grok. - It does make `Pri::from_str_variants()` redundant, while the `into_variant()` methods are refactored to `deserialize()` with a proper error message emitted to match the what serde would normally emit for failed enum variant deserialization. - A drawback of this change is that these two config fields lost the ability to reference a different value path in the `LogEvent`. That'll be addressed in a future commit.
In a YAML config a string can optionally be wrapped with quotes, while a number that isn't quote wrapped will be treated as a number type. The current support was only for string numbers, this change now supports flexibility for config using ordinal values in YAML regardless of quote usage. The previous `Self::into_variant(&s)` logic could have been used instead of bringing in `serde-aux`, but the external helper attribute approach seems easier to grok/follow as the intermediary container still seems required for a terse implementation. The match statement uses a reference (_which requires a deref for `from_repr`_) to appease the borrow checker for the later borrow needed by `value` in the error message.
This seems redundant given the context? Mostly adds unnecessary noise. Could probably `impl Configurable` or similar to try workaround the requirement. The metadata description could generate the variant list similar to how it's been handled for error message handling?
Not sure if this is worthwhile, but it adopts error message convention elsewhere I've seen by managing them via Snafu.
github-actions
bot
added
domain: sinks
Anything related to the Vector's sinks
domain: codecs
Anything related to Vector's codecs (encoding/decoding)
labels
Sep 18, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Labels
domain: codecs
Anything related to Vector's codecs (encoding/decoding)
domain: sinks
Anything related to the Vector's sinks
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Replaces: #17668
git blame
on my branch (Git web UI works fine for that) may provide the extra context needed 👍NOTE:
I'll try to allocate time within October to get those changes applied to this branch, along with a proper PR description at that point.
The conflicts were already resolved locally IIRC, they'll be taken care of with the future update.