Skip to content
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

Create span for plugin init #4472

Closed
BrynCooke opened this issue Jan 12, 2024 · 3 comments · Fixed by #4480
Closed

Create span for plugin init #4472

BrynCooke opened this issue Jan 12, 2024 · 3 comments · Fixed by #4480
Assignees

Comments

@BrynCooke
Copy link
Contributor

Plugin log messages should be correlated for initialization.
Suggest that a span is created around the call to plugin init that contains the following attributes:

Plugin name
A version supplied by the state machine (probably a UUID).
Doing this is a central location will give all plugins better contextual information on startup.

@Geal Geal self-assigned this Jan 15, 2024
@Geal
Copy link
Contributor

Geal commented Jan 15, 2024

so the interesting thing here is that right now we can't record spans happening on startup because the telemetry plugin is not created nor activated. It only works when reloading schema or config. On startup, we call the init_telemetry function, that creates a noop tracer:

pub(crate) fn init_telemetry(log_level: &str) -> Result<()> {
let hot_tracer = ReloadTracer::new(
opentelemetry::sdk::trace::TracerProvider::default().versioned_tracer(
"noop",
None::<String>,
None::<String>,
None,
),
);

the solution here could be to make a tracer instance that can record the spans, and once the telemetry plugin is created, we send the spans

@Geal
Copy link
Contributor

Geal commented Jan 15, 2024

or maybe, a better way would be to create and activate the telemetry plugin once, before everything is loaded

@Geal
Copy link
Contributor

Geal commented Jan 16, 2024

right, I tested in #4480 and it works well

Geal added a commit that referenced this issue Jan 24, 2024
Fix #4472 

This adds spans to create a trace during router initialization, at startup or during reloads. It will show how much time is spent initializing the query planner, each plugin and the request handling pipeline, and will be able to report events under those spans (ex: errors when intializing a plugin). That trace does not fall under normal sampling, it is always sent.

Co-authored-by: Coenen Benjamin <benjamin.coenen@hotmail.com>
This was referenced Feb 1, 2024
Geal added a commit that referenced this issue Feb 5, 2024
## 🚀 Features

### Specify Trace ID Formatting ([PR
#4530](#4530))

This adds the ability to specify the format of the trace ID in the
response headers of the supergraph service.

An example configuration making use of this feature is shown below:
```yaml
telemetry:
  apollo:
    client_name_header: name_header
    client_version_header: version_header
  exporters:
    tracing:
      experimental_response_trace_id:
        enabled: true
        header_name: trace_id
        format: decimal # Optional, defaults to hexadecimal
```

If the format is not specified, then the trace ID will continue to be in
hexadecimal format.

By [@nicholascioli](https://github.com/nicholascioli) in
#4530

### Introduce support for progressive `@override` ([PR
#4521](#4521))

The change brings support for progressive `@override`, which allows
dynamically overriding root fields and entity fields in the schema. This
feature is enterprise only and requires a license key to be used.

A new `label` argument is added to the `@override` directive in order to
indicate the field is dynamically overridden. Labels can come in two
forms:
1) String matching the form `percent(x)`: The router resolves these
labels based on the `x` value. For example, `percent(50)` will route 50%
of requests to the overridden field and 50% of requests to the original
field.
2) Arbitrary string matching the regex `^[a-zA-Z][a-zA-Z0-9_-:./]*$`:
These labels are expected to be resolved externally via coprocessor. A
coprocessor a supergraph request hook can inspect and modify the context
of a request in order to inform the router which labels to use during
query planning.

Please consult the docs for more information on how to use this feature
and how to implement a coprocessor for label resolution.

By [@trevorscheer](https://github.com/TrevorScheer) in
#4521

### Add a new selector to get all baggage key values in span attributes
([Issue #4425](#4425))

If you have several baggage items and would like to add all of them
directly as span attributes, for example `baggage: my_item=test,
my_second_item=bar` will add 2 span attributes `my_item=test` and
`my_second_item=bar`.

Example of configuration:

```yaml
telemetry:
  instrumentation:
    spans:
      router:
        attributes:
          baggage: true
```

By [@bnjjj](https://github.com/bnjjj) in
#4537

### Create a trace during router creation and plugin initialization
([Issue #4472](#4472))

When the router starts or reload, it will now generate a trace with
spans for query planner creation, schema parsing, plugin initialization
and request pipeline creation. This will help in debugging any issue
during startup, especially in plugins creation

By [@Geal](https://github.com/Geal) in
#4480

### Add static attribute on specific span in telemetry settings ([Issue
#4561](#4561))

Example of configuration:

```yaml
telemetry:
  instrumentation:
    spans:
      router:
        attributes:
          "my_attribute": "constant_value"
      supergraph:
        attributes:
          "my_attribute": "constant_value"
      subgraph:
        attributes:
          "my_attribute": "constant_value"
```

By [@bnjjj](https://github.com/bnjjj) in
#4566

## 🐛 Fixes

### Order HPA target so that kubernetes does not rewrite ([Issue
#4435](#4435))

This update addresses an OutOfSync issue in ArgoCD applications when
Horizontal Pod Autoscaler (HPA) is configured with both memory and CPU
limits.
Previously, the live and desired manifests within Kubernetes were not
consistently sorted, leading to persistent OutOfSync states in ArgoCD.
This change implements a sorting mechanism for HPA targets within the
Helm chart, ensuring alignment with Kubernetes' expected order.
This fix proactively resolves the sync discrepancies while using HPA,
circumventing the need to wait for Kubernetes' issue resolution
(kubernetes/kubernetes#74099).

By [@cyberhck](https://github.com/cyberhck) in
#4436

### Reactivate log events in traces ([PR
#4486](#4486))

This fixes a regression introduced in #2999, where events were not sent
with traces anymore due to too aggressive sampling

By [@Geal](https://github.com/Geal) in
#4486

### Fix inconsistency in environment variable parsing for telemetry
([Issue
#3203](https://github.com/apollographql/router/issues/ISSUE_NUMBER))

Previously, the router would complain when using the rover
recommendation of `APOLLO_TELEMETRY_DISABLED=1` environment
variable. Now any non-falsey value can be used, such as 1, yes, on,
etc..

By [@nicholascioli](https://github.com/nicholascioli) in
#4549

### Store static pages in `Bytes` structure to avoid expensive
allocation per request ([PR
#4528](#4528))

The `CheckpointService` created by the `StaticPageLayer` caused a
non-insignificant amount of memory to be allocated on every request. The
service stack gets cloned on every request, and so does the rendered
template.

The template is now stored in a `Bytes` struct instead which is cheap to
clone.

By [@xuorig](https://github.com/xuorig) in
#4528

### Fix header propagation issues ([Issue
#4312](#4312)), ([Issue
#4398](#4398))

This fixes two header propagation issues:
* if a client request header has already been added to a subgraph
request due to another header propagation rule, then it is only added
once
* `Accept`, `Accept-Encoding` and `Content-Encoding` were not in the
list of reserved headers that cannot be propagated. They are now in that
list because those headers are set explicitely by the Router in its
subgraph requests

There is a potential regression: if a router deployment was accidentally
relying on header propagation to compress subgraph requests, then it
will not work anymore because `Content-Encoding` is not propagated
anymore. Instead it should be set up from the `traffic_shaping` section
of the Router configuration:

```yaml
traffic_shaping:
  all:
    compression: gzip
  subgraphs: # Rules applied to requests from the router to individual subgraphs
    products:
      compression: identity
```

By [@Geal](https://github.com/Geal) in
#4535

## 🧪 Experimental

### Move cacheability metrics to the entity cache plugin ([Issue
#4253](#4253))

The metric was generated in the telemetry plugin before, but it was not
very convenient to keep it there. This adds more configuration:
- enable or disable the metrics
- set the metrics storage TTL (default is 60s)
- the metric's typename attribute is disabled by default. Activating it
can greatly increase the cardinality

This also includes some cleanup and performance improvements

By [@Geal](https://github.com/Geal) in
#4469

---------

Co-authored-by: Edward Huang <edward.huang@apollographql.com>
Co-authored-by: Jeremy Lempereur <jeremy.lempereur@iomentum.com>
Co-authored-by: Jesse Rosenberger <git@jro.cc>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging a pull request may close this issue.

2 participants