public/grafana
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

CloudMigrations: Query Grafana Migration Status for status while the snapshot is in the cloud (#90314)

Browse Source 
* implement querying gms for snapshot status

* add some documentation

* provide snapshot resources after snapshot is created

* add rate limiting to backend

* fix compilation error

* fix typo

* add unit tests

* finish merge

* lint

* swagger gen

* more testing

* remove duplicate test

* address a couple PR comments

* update switch statement to a map

* add timeouts to gms client through the http client

* remove extra whitespace

* put method back where it was so the PR is less confusing

* fix tests

* add todo

* fix final unit test
This commit is contained in:
Michael Mandrus
2024-07-15 09:22:57 -04:00
committed by GitHub
parent 5beaae8561
commit 542a1bf3ac
19 changed files with 532 additions and 125 deletions
Download Patch File Download Diff File Expand all files Collapse all files
pkg/services/cloudmigration/gmsclient/client.go
+1 -1
View File
@@ -10,7 +10,7 @@ type Client interface {
ValidateKey(context.Context, cloudmigration.CloudMigrationSession) error
MigrateData(context.Context, cloudmigration.CloudMigrationSession, cloudmigration.MigrateDataRequest) (*cloudmigration.MigrateDataResponse, error)
StartSnapshot(context.Context, cloudmigration.CloudMigrationSession) (*cloudmigration.StartSnapshotResponse, error)
GetSnapshotStatus(context.Context, cloudmigration.CloudMigrationSession, cloudmigration.CloudMigrationSnapshot) (*cloudmigration.CloudMigrationSnapshot, error)
GetSnapshotStatus(context.Context, cloudmigration.CloudMigrationSession, cloudmigration.CloudMigrationSnapshot) (*cloudmigration.GetSnapshotStatusResponse, error)
}
const logPrefix = "cloudmigration.gmsclient"
pkg/services/cloudmigration/gmsclient/dtos.go
-1
View File
@@ -1,4 +1,3 @@
// TODO: Move these to a shared library in common with GMS
package gmsclient
type MigrateDataType string
pkg/services/cloudmigration/gmsclient/gms_client.go
+76 -21
View File
@@ -9,29 +9,38 @@ import (
"io"
"net/http"
"strings"
"sync"
"time"
"github.com/grafana/grafana/pkg/infra/log"
"github.com/grafana/grafana/pkg/services/cloudmigration"
"github.com/grafana/grafana/pkg/setting"
)
// NewGMSClient returns an implementation of Client that queries GrafanaMigrationService
func NewGMSClient(domain string) Client {
return &gmsClientImpl{
domain: domain,
log: log.New(logPrefix),
func NewGMSClient(cfg *setting.Cfg) (Client, error) {
if cfg.CloudMigration.GMSDomain == "" {
return nil, fmt.Errorf("missing GMS domain")
}
return &gmsClientImpl{
cfg: cfg,
log: log.New(logPrefix),
}, nil
}
type gmsClientImpl struct {
domain string
log *log.ConcreteLogger
cfg *setting.Cfg
log *log.ConcreteLogger
getStatusMux sync.Mutex
getStatusLastQueried time.Time
}
func (c *gmsClientImpl) ValidateKey(ctx context.Context, cm cloudmigration.CloudMigrationSession) (err error) {
logger := c.log.FromContext(ctx)
// TODO update service url to gms
path := fmt.Sprintf("%s/api/v1/validate-key", buildBasePath(c.domain, cm.ClusterSlug))
// TODO: there is a lot of boilerplate code in these methods, we should consolidate them when we have a gardening period
path := fmt.Sprintf("%s/api/v1/validate-key", c.buildBasePath(cm.ClusterSlug))
// validation is an empty POST to GMS with the authorization header included
req, err := http.NewRequest("POST", path, bytes.NewReader(nil))
@@ -42,7 +51,9 @@ func (c *gmsClientImpl) ValidateKey(ctx context.Context, cm cloudmigration.Cloud
req.Header.Set("Content-Type", "application/json")
req.Header.Set("Authorization", fmt.Sprintf("Bearer %d:%s", cm.StackID, cm.AuthToken))
client := &http.Client{}
client := &http.Client{
Timeout: c.cfg.CloudMigration.GMSValidateKeyTimeout,
}
resp, err := client.Do(req)
if err != nil {
logger.Error("error sending http request for token validation", "err", err.Error())
@@ -62,11 +73,12 @@ func (c *gmsClientImpl) ValidateKey(ctx context.Context, cm cloudmigration.Cloud
return nil
}
// Deprecated
func (c *gmsClientImpl) MigrateData(ctx context.Context, cm cloudmigration.CloudMigrationSession, request cloudmigration.MigrateDataRequest) (result *cloudmigration.MigrateDataResponse, err error) {
logger := c.log.FromContext(ctx)
// TODO update service url to gms
path := fmt.Sprintf("%s/api/v1/migrate-data", buildBasePath(c.domain, cm.ClusterSlug))
path := fmt.Sprintf("%s/api/v1/migrate-data", c.buildBasePath(cm.ClusterSlug))
reqDTO := convertRequestToDTO(request)
body, err := json.Marshal(reqDTO)
@@ -111,7 +123,7 @@ func (c *gmsClientImpl) MigrateData(ctx context.Context, cm cloudmigration.Cloud
}
func (c *gmsClientImpl) StartSnapshot(ctx context.Context, session cloudmigration.CloudMigrationSession) (out *cloudmigration.StartSnapshotResponse, err error) {
path := fmt.Sprintf("%s/api/v1/start-snapshot", buildBasePath(c.domain, session.ClusterSlug))
path := fmt.Sprintf("%s/api/v1/start-snapshot", c.buildBasePath(session.ClusterSlug))
// Send the request to cms with the associated auth token
req, err := http.NewRequest(http.MethodPost, path, nil)
@@ -122,7 +134,9 @@ func (c *gmsClientImpl) StartSnapshot(ctx context.Context, session cloudmigratio
req.Header.Set("Content-Type", "application/json")
req.Header.Set("Authorization", fmt.Sprintf("Bearer %d:%s", session.StackID, session.AuthToken))
client := &http.Client{}
client := &http.Client{
Timeout: c.cfg.CloudMigration.GMSStartSnapshotTimeout,
}
resp, err := client.Do(req)
if err != nil {
c.log.Error("error sending http request to start snapshot", "err", err.Error())
@@ -148,8 +162,56 @@ func (c *gmsClientImpl) StartSnapshot(ctx context.Context, session cloudmigratio
return &result, nil
}
func (c *gmsClientImpl) GetSnapshotStatus(context.Context, cloudmigration.CloudMigrationSession, cloudmigration.CloudMigrationSnapshot) (*cloudmigration.CloudMigrationSnapshot, error) {
panic("not implemented")
func (c *gmsClientImpl) GetSnapshotStatus(ctx context.Context, session cloudmigration.CloudMigrationSession, snapshot cloudmigration.CloudMigrationSnapshot) (*cloudmigration.GetSnapshotStatusResponse, error) {
c.getStatusMux.Lock()
defer c.getStatusMux.Unlock()
logger := c.log.FromContext(ctx)
path := fmt.Sprintf("%s/api/v1/status/%s/status", c.buildBasePath(session.ClusterSlug), snapshot.GMSSnapshotUID)
// Send the request to gms with the associated auth token
req, err := http.NewRequest(http.MethodGet, path, nil)
if err != nil {
c.log.Error("error creating http request to get snapshot status", "err", err.Error())
return nil, fmt.Errorf("http request error: %w", err)
}
req.Header.Set("Content-Type", "application/json")
req.Header.Set("Authorization", fmt.Sprintf("Bearer %d:%s", session.StackID, session.AuthToken))
client := &http.Client{
Timeout: c.cfg.CloudMigration.GMSGetSnapshotStatusTimeout,
}
c.getStatusLastQueried = time.Now()
resp, err := client.Do(req)
if err != nil {
c.log.Error("error sending http request to get snapshot status", "err", err.Error())
return nil, fmt.Errorf("http request error: %w", err)
} else if resp.StatusCode >= 400 {
c.log.Error("received error response to get snapshot status", "statusCode", resp.StatusCode)
return nil, fmt.Errorf("http request error: %w", err)
}
defer func() {
if err := resp.Body.Close(); err != nil {
logger.Error("closing request body: %w", err)
}
}()
var result cloudmigration.GetSnapshotStatusResponse
if err := json.NewDecoder(resp.Body).Decode(&result); err != nil {
logger.Error("unmarshalling response body: %w", err)
return nil, fmt.Errorf("unmarshalling get snapshot status response: %w", err)
}
return &result, nil
}
func (c *gmsClientImpl) buildBasePath(clusterSlug string) string {
domain := c.cfg.CloudMigration.GMSDomain
if strings.HasPrefix(domain, "http://localhost") {
return domain
}
return fmt.Sprintf("https://cms-%s.%s/cloud-migrations", clusterSlug, domain)
}
func convertRequestToDTO(request cloudmigration.MigrateDataRequest) MigrateDataRequestDTO {
@@ -185,10 +247,3 @@ func convertResponseFromDTO(result MigrateDataResponseDTO) cloudmigration.Migrat
Items: items,
}
}
func buildBasePath(domain, clusterSlug string) string {
if strings.HasPrefix(domain, "http://localhost") {
return domain
}
return fmt.Sprintf("https://cms-%s.%s/cloud-migrations", clusterSlug, domain)
}
pkg/services/cloudmigration/gmsclient/gms_client_test.go
+21 -1
View File
@@ -3,12 +3,31 @@ package gmsclient
import (
"testing"
"github.com/grafana/grafana/pkg/setting"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
func Test_buildBasePath(t *testing.T) {
t.Parallel()
// Domain is required
_, err := NewGMSClient(&setting.Cfg{
CloudMigration: setting.CloudMigrationSettings{
GMSDomain: "",
},
})
require.Error(t, err)
// Domain is required
c, err := NewGMSClient(&setting.Cfg{
CloudMigration: setting.CloudMigrationSettings{
GMSDomain: "non-empty",
},
})
require.NoError(t, err)
client := c.(*gmsClientImpl)
tests := []struct {
description string
domain string
@@ -30,7 +49,8 @@ func Test_buildBasePath(t *testing.T) {
}
for _, tt := range tests {
t.Run(tt.description, func(t *testing.T) {
assert.Equal(t, tt.expected, buildBasePath(tt.domain, tt.clusterSlug))
client.cfg.CloudMigration.GMSDomain = tt.domain
assert.Equal(t, tt.expected, client.buildBasePath(tt.clusterSlug))
})
}
}
pkg/services/cloudmigration/gmsclient/inmemory_client.go
+21 -27
View File
@@ -4,7 +4,6 @@ import (
"context"
"fmt"
"math/rand"
"time"
cryptoRand "crypto/rand"
@@ -68,33 +67,28 @@ func (c *memoryClientImpl) StartSnapshot(context.Context, cloudmigration.CloudMi
return c.snapshot, nil
}
func (c *memoryClientImpl) GetSnapshotStatus(ctx context.Context, session cloudmigration.CloudMigrationSession, snapshot cloudmigration.CloudMigrationSnapshot) (*cloudmigration.CloudMigrationSnapshot, error) {
results := []cloudmigration.CloudMigrationResource{
{
Type: cloudmigration.DashboardDataType,
RefID: "dash1",
Status: cloudmigration.ItemStatusOK,
},
{
Type: cloudmigration.DatasourceDataType,
RefID: "ds1",
Status: cloudmigration.ItemStatusError,
Error: "fake error",
},
{
Type: cloudmigration.FolderDataType,
RefID: "folder1",
Status: cloudmigration.ItemStatusOK,
func (c *memoryClientImpl) GetSnapshotStatus(ctx context.Context, session cloudmigration.CloudMigrationSession, snapshot cloudmigration.CloudMigrationSnapshot) (*cloudmigration.GetSnapshotStatusResponse, error) {
gmsResp := &cloudmigration.GetSnapshotStatusResponse{
State: cloudmigration.SnapshotStateFinished,
Results: []cloudmigration.CloudMigrationResource{
{
Type: cloudmigration.DashboardDataType,
RefID: "dash1",
Status: cloudmigration.ItemStatusOK,
},
{
Type: cloudmigration.DatasourceDataType,
RefID: "ds1",
Status: cloudmigration.ItemStatusError,
Error: "fake error",
},
{
Type: cloudmigration.FolderDataType,
RefID: "folder1",
Status: cloudmigration.ItemStatusOK,
},
},
}
// just fake an entire response
gmsSnapshot := cloudmigration.CloudMigrationSnapshot{
Status: cloudmigration.SnapshotStatusFinished,
GMSSnapshotUID: "gmssnapshotuid",
Resources: results,
Finished: time.Now(),
}
return &gmsSnapshot, nil
return gmsResp, nil
}