cloudwatch.go 9.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392
  1. package cloudwatch
  2. import (
  3. "context"
  4. "errors"
  5. "fmt"
  6. "regexp"
  7. "sort"
  8. "strconv"
  9. "strings"
  10. "time"
  11. "github.com/grafana/grafana/pkg/log"
  12. "github.com/grafana/grafana/pkg/models"
  13. "github.com/grafana/grafana/pkg/setting"
  14. "github.com/grafana/grafana/pkg/tsdb"
  15. "github.com/aws/aws-sdk-go/aws"
  16. "github.com/aws/aws-sdk-go/aws/request"
  17. "github.com/aws/aws-sdk-go/service/cloudwatch"
  18. "github.com/aws/aws-sdk-go/service/ec2/ec2iface"
  19. "github.com/grafana/grafana/pkg/components/null"
  20. "github.com/grafana/grafana/pkg/components/simplejson"
  21. "github.com/grafana/grafana/pkg/metrics"
  22. )
  23. type CloudWatchExecutor struct {
  24. *models.DataSource
  25. ec2Svc ec2iface.EC2API
  26. }
  27. type DatasourceInfo struct {
  28. Profile string
  29. Region string
  30. AuthType string
  31. AssumeRoleArn string
  32. Namespace string
  33. AccessKey string
  34. SecretKey string
  35. }
  36. func NewCloudWatchExecutor(dsInfo *models.DataSource) (tsdb.TsdbQueryEndpoint, error) {
  37. return &CloudWatchExecutor{}, nil
  38. }
  39. var (
  40. plog log.Logger
  41. standardStatistics map[string]bool
  42. aliasFormat *regexp.Regexp
  43. )
  44. func init() {
  45. plog = log.New("tsdb.cloudwatch")
  46. tsdb.RegisterTsdbQueryEndpoint("cloudwatch", NewCloudWatchExecutor)
  47. standardStatistics = map[string]bool{
  48. "Average": true,
  49. "Maximum": true,
  50. "Minimum": true,
  51. "Sum": true,
  52. "SampleCount": true,
  53. }
  54. aliasFormat = regexp.MustCompile(`\{\{\s*(.+?)\s*\}\}`)
  55. }
  56. func (e *CloudWatchExecutor) Query(ctx context.Context, dsInfo *models.DataSource, queryContext *tsdb.TsdbQuery) (*tsdb.Response, error) {
  57. var result *tsdb.Response
  58. e.DataSource = dsInfo
  59. queryType := queryContext.Queries[0].Model.Get("type").MustString("")
  60. var err error
  61. switch queryType {
  62. case "metricFindQuery":
  63. result, err = e.executeMetricFindQuery(ctx, queryContext)
  64. case "annotationQuery":
  65. result, err = e.executeAnnotationQuery(ctx, queryContext)
  66. case "timeSeriesQuery":
  67. fallthrough
  68. default:
  69. result, err = e.executeTimeSeriesQuery(ctx, queryContext)
  70. }
  71. return result, err
  72. }
  73. func (e *CloudWatchExecutor) executeTimeSeriesQuery(ctx context.Context, queryContext *tsdb.TsdbQuery) (*tsdb.Response, error) {
  74. result := &tsdb.Response{
  75. Results: make(map[string]*tsdb.QueryResult),
  76. }
  77. errCh := make(chan error, 1)
  78. resCh := make(chan *tsdb.QueryResult, 1)
  79. currentlyExecuting := 0
  80. for i, model := range queryContext.Queries {
  81. queryType := model.Model.Get("type").MustString()
  82. if queryType != "timeSeriesQuery" && queryType != "" {
  83. continue
  84. }
  85. currentlyExecuting++
  86. go func(refId string, index int) {
  87. queryRes, err := e.executeQuery(ctx, queryContext.Queries[index].Model, queryContext)
  88. currentlyExecuting--
  89. if err != nil {
  90. errCh <- err
  91. } else {
  92. queryRes.RefId = refId
  93. resCh <- queryRes
  94. }
  95. }(model.RefId, i)
  96. }
  97. for currentlyExecuting != 0 {
  98. select {
  99. case res := <-resCh:
  100. result.Results[res.RefId] = res
  101. case err := <-errCh:
  102. return result, err
  103. case <-ctx.Done():
  104. return result, ctx.Err()
  105. }
  106. }
  107. return result, nil
  108. }
  109. func (e *CloudWatchExecutor) executeQuery(ctx context.Context, parameters *simplejson.Json, queryContext *tsdb.TsdbQuery) (*tsdb.QueryResult, error) {
  110. query, err := parseQuery(parameters)
  111. if err != nil {
  112. return nil, err
  113. }
  114. client, err := e.getClient(query.Region)
  115. if err != nil {
  116. return nil, err
  117. }
  118. startTime, err := queryContext.TimeRange.ParseFrom()
  119. if err != nil {
  120. return nil, err
  121. }
  122. endTime, err := queryContext.TimeRange.ParseTo()
  123. if err != nil {
  124. return nil, err
  125. }
  126. if endTime.Before(startTime) {
  127. return nil, fmt.Errorf("Invalid time range: End time can't be before start time")
  128. }
  129. params := &cloudwatch.GetMetricStatisticsInput{
  130. Namespace: aws.String(query.Namespace),
  131. MetricName: aws.String(query.MetricName),
  132. Dimensions: query.Dimensions,
  133. Period: aws.Int64(int64(query.Period)),
  134. }
  135. if len(query.Statistics) > 0 {
  136. params.Statistics = query.Statistics
  137. }
  138. if len(query.ExtendedStatistics) > 0 {
  139. params.ExtendedStatistics = query.ExtendedStatistics
  140. }
  141. // 1 minutes resolutin metrics is stored for 15 days, 15 * 24 * 60 = 21600
  142. if query.HighResolution && (((endTime.Unix() - startTime.Unix()) / int64(query.Period)) > 21600) {
  143. return nil, errors.New("too long query period")
  144. }
  145. var resp *cloudwatch.GetMetricStatisticsOutput
  146. for startTime.Before(endTime) {
  147. params.StartTime = aws.Time(startTime)
  148. if query.HighResolution {
  149. startTime = startTime.Add(time.Duration(1440*query.Period) * time.Second)
  150. } else {
  151. startTime = endTime
  152. }
  153. params.EndTime = aws.Time(startTime)
  154. if setting.Env == setting.DEV {
  155. plog.Debug("CloudWatch query", "raw query", params)
  156. }
  157. partResp, err := client.GetMetricStatisticsWithContext(ctx, params, request.WithResponseReadTimeout(10*time.Second))
  158. if err != nil {
  159. return nil, err
  160. }
  161. if resp != nil {
  162. resp.Datapoints = append(resp.Datapoints, partResp.Datapoints...)
  163. } else {
  164. resp = partResp
  165. }
  166. metrics.M_Aws_CloudWatch_GetMetricStatistics.Inc()
  167. }
  168. queryRes, err := parseResponse(resp, query)
  169. if err != nil {
  170. return nil, err
  171. }
  172. return queryRes, nil
  173. }
  174. func parseDimensions(model *simplejson.Json) ([]*cloudwatch.Dimension, error) {
  175. var result []*cloudwatch.Dimension
  176. for k, v := range model.Get("dimensions").MustMap() {
  177. kk := k
  178. if vv, ok := v.(string); ok {
  179. result = append(result, &cloudwatch.Dimension{
  180. Name: &kk,
  181. Value: &vv,
  182. })
  183. } else {
  184. return nil, errors.New("failed to parse")
  185. }
  186. }
  187. sort.Slice(result, func(i, j int) bool {
  188. return *result[i].Name < *result[j].Name
  189. })
  190. return result, nil
  191. }
  192. func parseStatistics(model *simplejson.Json) ([]string, []string, error) {
  193. var statistics []string
  194. var extendedStatistics []string
  195. for _, s := range model.Get("statistics").MustArray() {
  196. if ss, ok := s.(string); ok {
  197. if _, isStandard := standardStatistics[ss]; isStandard {
  198. statistics = append(statistics, ss)
  199. } else {
  200. extendedStatistics = append(extendedStatistics, ss)
  201. }
  202. } else {
  203. return nil, nil, errors.New("failed to parse")
  204. }
  205. }
  206. return statistics, extendedStatistics, nil
  207. }
  208. func parseQuery(model *simplejson.Json) (*CloudWatchQuery, error) {
  209. region, err := model.Get("region").String()
  210. if err != nil {
  211. return nil, err
  212. }
  213. namespace, err := model.Get("namespace").String()
  214. if err != nil {
  215. return nil, err
  216. }
  217. metricName, err := model.Get("metricName").String()
  218. if err != nil {
  219. return nil, err
  220. }
  221. dimensions, err := parseDimensions(model)
  222. if err != nil {
  223. return nil, err
  224. }
  225. statistics, extendedStatistics, err := parseStatistics(model)
  226. if err != nil {
  227. return nil, err
  228. }
  229. p := model.Get("period").MustString("")
  230. if p == "" {
  231. if namespace == "AWS/EC2" {
  232. p = "300"
  233. } else {
  234. p = "60"
  235. }
  236. }
  237. var period int
  238. if regexp.MustCompile(`^\d+$`).Match([]byte(p)) {
  239. period, err = strconv.Atoi(p)
  240. if err != nil {
  241. return nil, err
  242. }
  243. } else {
  244. d, err := time.ParseDuration(p)
  245. if err != nil {
  246. return nil, err
  247. }
  248. period = int(d.Seconds())
  249. }
  250. alias := model.Get("alias").MustString()
  251. if alias == "" {
  252. alias = "{{metric}}_{{stat}}"
  253. }
  254. highResolution := model.Get("highResolution").MustBool(false)
  255. return &CloudWatchQuery{
  256. Region: region,
  257. Namespace: namespace,
  258. MetricName: metricName,
  259. Dimensions: dimensions,
  260. Statistics: aws.StringSlice(statistics),
  261. ExtendedStatistics: aws.StringSlice(extendedStatistics),
  262. Period: period,
  263. Alias: alias,
  264. HighResolution: highResolution,
  265. }, nil
  266. }
  267. func formatAlias(query *CloudWatchQuery, stat string, dimensions map[string]string) string {
  268. data := map[string]string{}
  269. data["region"] = query.Region
  270. data["namespace"] = query.Namespace
  271. data["metric"] = query.MetricName
  272. data["stat"] = stat
  273. data["period"] = strconv.Itoa(query.Period)
  274. for k, v := range dimensions {
  275. data[k] = v
  276. }
  277. result := aliasFormat.ReplaceAllFunc([]byte(query.Alias), func(in []byte) []byte {
  278. labelName := strings.Replace(string(in), "{{", "", 1)
  279. labelName = strings.Replace(labelName, "}}", "", 1)
  280. labelName = strings.TrimSpace(labelName)
  281. if val, exists := data[labelName]; exists {
  282. return []byte(val)
  283. }
  284. return in
  285. })
  286. return string(result)
  287. }
  288. func parseResponse(resp *cloudwatch.GetMetricStatisticsOutput, query *CloudWatchQuery) (*tsdb.QueryResult, error) {
  289. queryRes := tsdb.NewQueryResult()
  290. var value float64
  291. for _, s := range append(query.Statistics, query.ExtendedStatistics...) {
  292. series := tsdb.TimeSeries{
  293. Tags: map[string]string{},
  294. Points: make([]tsdb.TimePoint, 0),
  295. }
  296. for _, d := range query.Dimensions {
  297. series.Tags[*d.Name] = *d.Value
  298. }
  299. series.Name = formatAlias(query, *s, series.Tags)
  300. lastTimestamp := make(map[string]time.Time)
  301. sort.Slice(resp.Datapoints, func(i, j int) bool {
  302. return (*resp.Datapoints[i].Timestamp).Before(*resp.Datapoints[j].Timestamp)
  303. })
  304. for _, v := range resp.Datapoints {
  305. switch *s {
  306. case "Average":
  307. value = *v.Average
  308. case "Maximum":
  309. value = *v.Maximum
  310. case "Minimum":
  311. value = *v.Minimum
  312. case "Sum":
  313. value = *v.Sum
  314. case "SampleCount":
  315. value = *v.SampleCount
  316. default:
  317. if strings.Index(*s, "p") == 0 && v.ExtendedStatistics[*s] != nil {
  318. value = *v.ExtendedStatistics[*s]
  319. }
  320. }
  321. // terminate gap of data points
  322. timestamp := *v.Timestamp
  323. if _, ok := lastTimestamp[*s]; ok {
  324. nextTimestampFromLast := lastTimestamp[*s].Add(time.Duration(query.Period) * time.Second)
  325. for timestamp.After(nextTimestampFromLast) {
  326. series.Points = append(series.Points, tsdb.NewTimePoint(null.FloatFromPtr(nil), float64(nextTimestampFromLast.Unix()*1000)))
  327. nextTimestampFromLast = nextTimestampFromLast.Add(time.Duration(query.Period) * time.Second)
  328. }
  329. }
  330. lastTimestamp[*s] = timestamp
  331. series.Points = append(series.Points, tsdb.NewTimePoint(null.FloatFrom(value), float64(timestamp.Unix()*1000)))
  332. }
  333. queryRes.Series = append(queryRes.Series, &series)
  334. }
  335. return queryRes, nil
  336. }