-
Notifications
You must be signed in to change notification settings - Fork 42
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
[Fix] Support custom retry logic per method #1081
base: main
Are you sure you want to change the base?
Conversation
…ent" requests varies somewhat from service to service
@@ -63,8 +63,7 @@ func (c *DatabricksClient) GetOAuthToken(ctx context.Context, authDetails string | |||
|
|||
// Do sends an HTTP request against path. | |||
func (c *DatabricksClient) Do(ctx context.Context, method, path string, | |||
headers map[string]string, request, response any, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I can add this back in if desired, just a small formatting change.
} | ||
return false | ||
}, | ||
httpclient.RetryUrlErrors, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I moved this out of the ApiClient to have a single codesite where the retry logic is defined for the client. The downside is that you need to add this explicitly in your ErrorRetriable if you don't specify DefaultErrorRetriable. Happy to make this a default behavior, let me know what you think.
Test Details: go/deco-tests/11666553890 |
client/client.go
Outdated
@@ -63,8 +63,7 @@ func (c *DatabricksClient) GetOAuthToken(ctx context.Context, authDetails string | |||
|
|||
// Do sends an HTTP request against path. | |||
func (c *DatabricksClient) Do(ctx context.Context, method, path string, | |||
headers map[string]string, request, response any, | |||
visitors ...func(*http.Request) error) error { | |||
headers map[string]string, request, response any, visitors ...func(*http.Request) error) error { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
[optional] I would go with one parameter per line to echo the way function calls and struct declarations are made:
func (c *DatabricksClient) Do(
ctx context.Context,
method string,
path string,
headers map[string]string,
request any,
response any,
visitors ...func(*http.Request) error
) error {
There's a couple of similar patterns in the Go standard library but not many. One of the reason is that long lists of parameters are usually substituted with a struct (https://google.github.io/styleguide/go/best-practices#option-structure). I actually wanted to make that change for quite sometime but it didn't feel right sending one PR just for that.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yeah. This Do()
method should be formulated either as a struct argument or with functional options, since everything after path
is optional. I'd prefer not to change this signature unless necessary, as it is used in multiple places in the TF provider.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Agreed with not using a struct as this is out of scope for this PR. Let's at least format idiomatically then. Our Go style does not mandate a 80 char line length which makes the current formatting quite arbitrary. I'm fine with either having everything on a single line or one argument per line.
package config | ||
|
||
import ( | ||
"context" | ||
"fmt" | ||
"io" | ||
"net/http" | ||
"strings" | ||
"testing" | ||
|
||
"github.com/databricks/databricks-sdk-go/httpclient" | ||
"github.com/stretchr/testify/assert" | ||
"github.com/stretchr/testify/require" | ||
) | ||
|
||
type hc func(r *http.Request) (*http.Response, error) | ||
|
||
func (cb hc) RoundTrip(r *http.Request) (*http.Response, error) { | ||
return cb(r) | ||
} | ||
|
||
func (cb hc) SkipRetryOnIO() bool { | ||
return true | ||
} | ||
|
||
func TestApiClient_RetriesGetPermissionsOnGatewayTimeout(t *testing.T) { | ||
requestCount := 0 | ||
c := &Config{ | ||
HTTPTransport: hc(func(r *http.Request) (*http.Response, error) { | ||
initialRequestCount := requestCount | ||
requestCount++ | ||
if initialRequestCount == 0 { | ||
return &http.Response{ | ||
Request: r, | ||
StatusCode: http.StatusGatewayTimeout, | ||
Body: io.NopCloser(strings.NewReader( | ||
fmt.Sprintf(`{"error_code":"TEMPORARILY_UNAVAILABLE", "message":"The service at %s is taking too long to process your request. Please try again later or try a faster operation."}`, r.URL))), | ||
}, nil | ||
} | ||
return &http.Response{ | ||
Request: r, | ||
StatusCode: http.StatusOK, | ||
Body: io.NopCloser(strings.NewReader(`{"permissions": ["can_run_queries"]}`)), | ||
}, nil | ||
}), | ||
} | ||
client, err := c.NewApiClient() | ||
require.NoError(t, err) | ||
ctx := context.Background() | ||
var res map[string][]string | ||
err = client.Do(ctx, "GET", "/api/2.0/permissions/object/id", httpclient.WithResponseUnmarshal(&res)) | ||
assert.NoError(t, err) | ||
assert.Equal(t, map[string][]string{"permissions": {"can_run_queries"}}, res) | ||
} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'd recommend structuring the unit tests differently by having this test focused on how ApiClient
manages mocked ErrorRetrier
(or the absence of an ErrorRetrier
). The tests of the ErrorRetrier
themselves (e.g. verify that the path pattern match properly) should happen in httpclient/errors_test.go
.
This test could look like the following (I did not verify that the code works):
type mock struct {
MaxFails int // number of times the failed Response is returned
FailResponse *http.Response // response to return in case of fail
FailError error // error to return in case of fail
NumCalls int // total number of calls
}
func (m *mock) RoundTrip(r *http.Request) (*http.Response, error) {
m.NumCalls++
if m.NumCalls <= m.MaxFails {
return m.FailResponse, n.FailError
}
return &http.Response{
Request: r,
StatusCode: http.StatusOK,
Body: io.NopCloser(strings.NewReader(`{}`)),
}, nil
}
func (m *mock) SkipRetryOnIO() bool {
return true
}
func TestApiClient_Do_retries(t *testing.T) {
testCases := []struct{
desc string
config *Config
errorRetrier ErrorRetrier
wantNumCalls int
} {
{
desc: "nil retrier",
mock: &mock{
MaxFails: 1,
FailResponse: &http.Response{StatusCode: http.StatusGatewayTimeout}
}
wantNumCalls: 1,
},
{
desc: "no retry",
mock: &mock{
MaxFails: 1,
FailResponse: &http.Response{StatusCode: http.StatusGatewayTimeout}
}
errorRetrier: func(context.Context, *http.Request, *common.ResponseWrapper, error) bool {
return false
},
wantNumCalls: 1,
},
{
desc: "retry 1 time",
mock: &mock{
MaxFails: 1,
FailResponse: &http.Response{StatusCode: http.StatusGatewayTimeout}
}
errorRetrier: func(context.Context, *http.Request, *common.ResponseWrapper, error) bool {
return true
},
wantNumCalls: 2,
},
{
desc: "retry 2 times",
mock: &mock{
MaxFails: 2,
FailResponse: &http.Response{StatusCode: http.StatusGatewayTimeout}
}
errorRetrier: func(_ context.Context, _ *http.Request, _ *common.ResponseWrapper, _ error) bool {
return true
},
wantNumCalls: 3,
},
{
desc: "retry 3 times",
mock: &mock{
MaxFails: 3,
FailResponse: &http.Response{StatusCode: http.StatusGatewayTimeout}
}
errorRetrier: func(_ context.Context, _ *http.Request, _ *common.ResponseWrapper, _ error) bool {
return true
},
wantNumCalls: 4,
},
}
func _, tc := range testCases {
t.Run(tc.desc, func(t *testing.T) {
cfg := &Config{HTTPTransport: tc.mock}
client, err := cfg.NewApiClient()
client.ErrorRetrier = tc.errorRetrier
err = client.Do(context.Background(), "GET", "test-path")
gotNumCalls = tc.mock.NumCalls
if gotNumCalls != tc.wantNumCalls {
t.Errorf("got %d calls, want %d", gotNumCalls, tc.wantNumCalls)
}
})
}
}
Please feel free to ignore this comment if this is too much work or if the ApiClient cannot be instrumented that easily.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It definitely can be instrumented this way, and this is a nice test case to use (I'll adapt it and include it in this PR). However, I did want to specifically test the get permissions pathway. Essentially, this tests that "the client returned by Config.GetApiClient() correctly implements retry on 504." I will add more test cases here though.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I did want to specifically test the get permissions pathway
Sounds good to me as long as this complements the overall testing of the retry logic.
If integration tests don't run automatically, an authorized user can run them manually by following the instructions below: Trigger: Inputs:
Checks will be approved automatically on success. |
RetryTimeout time.Duration | ||
|
||
// Returns the amount of time to wait after the given attempt. | ||
RetryBackoff retries.BackoffFunc |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This is added here and exposed initially for testing, though it should be safe for users to configure this if they need to modify the backoff behavior.
Changes
Some requests can be safely retried when the API returns a 504 Gateway Timeout, but not all. In this PR, we refactor the retry logic slightly to allow discriminating based on the request and response as well as any potential
error
returned from the Go HTTP client or parsed from the API response. The logic for this is consolidated into a single place, theErrorRetriable
field of the HTTP Client's config.In addition, we add retries for the Get Permissions API, which can fail in this way, to provide better stability for callers of this endpoint. Today, we don't have a clear indication of which API methods are idempotent, so the approach is very conservative, only retrying requests whose method and path match a given pattern.
This PR incidentally rewrites the retry logic for the DefaultErrorRetriable, reading the status code directly from the
*http.Response
instead of from the parsed error. This is less brittle in case a custom error parser is provided that doesn't deserialize 4xx/5xx responses using theHttpError
type.Tests
The first commit is a refactor of the existing code. All existing tests pass.
A new test is added that verifies that the Get Permissions API retries when the API responds with a 504. I verified that this fails on the initial commit.
make test
passingmake fmt
applied