ruler.go 16.7 KB
Newer Older
J
junotx 已提交
1 2 3 4 5
package rules

import (
	"context"
	"fmt"
J
junotx 已提交
6
	"net/http"
J
junotx 已提交
7 8
	"sort"

J
junotx 已提交
9
	"github.com/docker/docker/pkg/locker"
J
junotx 已提交
10 11 12 13 14 15
	"github.com/ghodss/yaml"
	"github.com/pkg/errors"
	promresourcesv1 "github.com/prometheus-operator/prometheus-operator/pkg/apis/monitoring/v1"
	prominformersv1 "github.com/prometheus-operator/prometheus-operator/pkg/client/informers/externalversions/monitoring/v1"
	promresourcesclient "github.com/prometheus-operator/prometheus-operator/pkg/client/versioned"
	corev1 "k8s.io/api/core/v1"
J
junotx 已提交
16
	apierrors "k8s.io/apimachinery/pkg/api/errors"
J
junotx 已提交
17 18
	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
	"k8s.io/apimachinery/pkg/labels"
J
junotx 已提交
19
	"k8s.io/client-go/util/retry"
20
	"kubesphere.io/kubesphere/pkg/api/alerting/v2alpha1"
J
junotx 已提交
21 22 23 24 25 26 27 28
)

const (
	customAlertingRuleResourcePrefix = "custom-alerting-rule-"
)

var (
	maxSecretSize        = corev1.MaxSecretSize
J
junotx 已提交
29
	maxConfigMapDataSize = int(float64(maxSecretSize) * 0.45)
J
junotx 已提交
30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130

	errOutOfConfigMapSize = errors.New("out of config map size")
)

type Ruler interface {
	Namespace() string
	RuleResourceNamespaceSelector() (labels.Selector, error)
	RuleResourceSelector(extraRuleResourceSelector labels.Selector) (labels.Selector, error)
	ExternalLabels() func() map[string]string

	ListRuleResources(ruleNamespace *corev1.Namespace, extraRuleResourceSelector labels.Selector) (
		[]*promresourcesv1.PrometheusRule, error)

	AddAlertingRule(ctx context.Context, ruleNamespace *corev1.Namespace, extraRuleResourceSelector labels.Selector,
		group string, rule *promresourcesv1.Rule, ruleResourceLabels map[string]string) error
	UpdateAlertingRule(ctx context.Context, ruleNamespace *corev1.Namespace, extraRuleResourceSelector labels.Selector,
		group string, rule *promresourcesv1.Rule, ruleResourceLabels map[string]string) error
	DeleteAlertingRule(ctx context.Context, ruleNamespace *corev1.Namespace, extraRuleResourceSelector labels.Selector,
		group string, name string) error
}

type ruleResource promresourcesv1.PrometheusRule

// deleteAlertingRule deletes the rules with the given name.
// If the rule is deleted, return true to indicate the resource should be updated.
func (r *ruleResource) deleteAlertingRule(name string) (bool, error) {
	var (
		nGroups []promresourcesv1.RuleGroup
		ok      bool
	)

	for _, g := range r.Spec.Groups {
		var rules []promresourcesv1.Rule
		for _, gr := range g.Rules {
			if gr.Alert != "" && gr.Alert == name {
				ok = true
				continue
			}
			rules = append(rules, gr)
		}
		if len(rules) > 0 {
			nGroups = append(nGroups, promresourcesv1.RuleGroup{
				Name:                    g.Name,
				Interval:                g.Interval,
				PartialResponseStrategy: g.PartialResponseStrategy,
				Rules:                   rules,
			})
		}
	}

	if ok {
		r.Spec.Groups = nGroups
	}
	return ok, nil
}

// updateAlertingRule updates the rule with the given group.
// If the rule is updated, return true to indicate the resource should be updated.
func (r *ruleResource) updateAlertingRule(groupName string, rule *promresourcesv1.Rule) (bool, error) {
	var (
		ok       bool
		pr       = (promresourcesv1.PrometheusRule)(*r)
		npr      = pr.DeepCopy()
		groupMap = make(map[string]*promresourcesv1.RuleGroup)
	)

	for _, g := range npr.Spec.Groups {
		var rules []promresourcesv1.Rule
		for i, gr := range g.Rules {
			if gr.Alert != "" && gr.Alert == rule.Alert {
				ok = true
				continue
			}
			rules = append(rules, g.Rules[i])
		}
		if len(rules) > 0 {
			groupMap[g.Name] = &promresourcesv1.RuleGroup{
				Name:                    g.Name,
				Interval:                g.Interval,
				PartialResponseStrategy: g.PartialResponseStrategy,
				Rules:                   rules,
			}
		}
	}

	if ok {
		if g, exist := groupMap[groupName]; exist {
			g.Rules = append(g.Rules, *rule)
		} else {
			groupMap[groupName] = &promresourcesv1.RuleGroup{
				Name:  groupName,
				Rules: []promresourcesv1.Rule{*rule},
			}
		}

		var groups []promresourcesv1.RuleGroup
		for _, g := range groupMap {
			groups = append(groups, *g)
		}

		npr.Spec.Groups = groups
J
junotx 已提交
131
		content, err := yaml.Marshal(npr.Spec)
J
junotx 已提交
132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166
		if err != nil {
			return false, errors.Wrap(err, "failed to unmarshal content")
		}

		if len(string(content)) < maxConfigMapDataSize { // check size limit
			r.Spec.Groups = groups
			return true, nil
		}
		return false, errOutOfConfigMapSize
	}
	return false, nil
}

func (r *ruleResource) addAlertingRule(group string, rule *promresourcesv1.Rule) (bool, error) {
	var (
		err error
		pr  = (promresourcesv1.PrometheusRule)(*r)
		npr = pr.DeepCopy()
		ok  bool
	)

	for i := 0; i < len(npr.Spec.Groups); i++ {
		if npr.Spec.Groups[i].Name == group {
			npr.Spec.Groups[i].Rules = append(npr.Spec.Groups[i].Rules, *rule)
			ok = true
			break
		}
	}
	if !ok { // add a group when there is no group with the specified group name
		npr.Spec.Groups = append(npr.Spec.Groups, promresourcesv1.RuleGroup{
			Name:  group,
			Rules: []promresourcesv1.Rule{*rule},
		})
	}

J
junotx 已提交
167
	content, err := yaml.Marshal(npr.Spec)
J
junotx 已提交
168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274
	if err != nil {
		return false, errors.Wrap(err, "failed to unmarshal content")
	}

	if len(string(content)) < maxConfigMapDataSize { // check size limit
		r.Spec.Groups = npr.Spec.Groups
		return true, nil
	} else {
		return false, errOutOfConfigMapSize
	}
}

func (r *ruleResource) commit(ctx context.Context, prometheusResourceClient promresourcesclient.Interface) error {
	var pr = (promresourcesv1.PrometheusRule)(*r)
	if len(pr.Spec.Groups) == 0 {
		return prometheusResourceClient.MonitoringV1().PrometheusRules(r.Namespace).Delete(ctx, r.Name, metav1.DeleteOptions{})
	}
	newPr, err := prometheusResourceClient.MonitoringV1().PrometheusRules(r.Namespace).Update(ctx, &pr, metav1.UpdateOptions{})
	if err != nil {
		return err
	}
	newPr.DeepCopyInto(&pr)
	return nil
}

type PrometheusRuler struct {
	resource *promresourcesv1.Prometheus
	informer prominformersv1.PrometheusRuleInformer
	client   promresourcesclient.Interface
}

func NewPrometheusRuler(resource *promresourcesv1.Prometheus, informer prominformersv1.PrometheusRuleInformer,
	client promresourcesclient.Interface) Ruler {
	return &PrometheusRuler{
		resource: resource,
		informer: informer,
		client:   client,
	}
}

func (r *PrometheusRuler) Namespace() string {
	return r.resource.Namespace
}

func (r *PrometheusRuler) RuleResourceNamespaceSelector() (labels.Selector, error) {
	if r.resource.Spec.RuleNamespaceSelector == nil {
		return nil, nil
	}
	return metav1.LabelSelectorAsSelector(r.resource.Spec.RuleNamespaceSelector)
}

func (r *PrometheusRuler) RuleResourceSelector(extraRuleResourceSelector labels.Selector) (labels.Selector, error) {
	rSelector, err := metav1.LabelSelectorAsSelector(r.resource.Spec.RuleSelector)
	if err != nil {
		return nil, err
	}
	if extraRuleResourceSelector != nil {
		if requirements, ok := extraRuleResourceSelector.Requirements(); ok {
			rSelector = rSelector.Add(requirements...)
		}
	}
	return rSelector, nil
}

func (r *PrometheusRuler) ExternalLabels() func() map[string]string {
	// ignoring the external labels because rules gotten from prometheus endpoint do not include them
	return nil
}

func (r *PrometheusRuler) ListRuleResources(ruleNamespace *corev1.Namespace, extraRuleResourceSelector labels.Selector) (
	[]*promresourcesv1.PrometheusRule, error) {
	selected, err := ruleNamespaceSelected(r, ruleNamespace)
	if err != nil {
		return nil, err
	}
	if !selected {
		return nil, nil
	}
	rSelector, err := r.RuleResourceSelector(extraRuleResourceSelector)
	if err != nil {
		return nil, err
	}
	return r.informer.Lister().PrometheusRules(ruleNamespace.Name).List(rSelector)
}

func (r *PrometheusRuler) AddAlertingRule(ctx context.Context, ruleNamespace *corev1.Namespace,
	extraRuleResourceSelector labels.Selector,
	group string, rule *promresourcesv1.Rule, ruleResourceLabels map[string]string) error {
	return errors.New("not supported to add rules for prometheus")
}

func (r *PrometheusRuler) UpdateAlertingRule(ctx context.Context, ruleNamespace *corev1.Namespace,
	extraRuleResourceSelector labels.Selector,
	group string, rule *promresourcesv1.Rule, ruleResourceLabels map[string]string) error {
	return errors.New("not supported to update rules for prometheus")
}

func (r *PrometheusRuler) DeleteAlertingRule(ctx context.Context, ruleNamespace *corev1.Namespace,
	extraRuleResourceSelector labels.Selector,
	group string, name string) error {
	return errors.New("not supported to update rules for prometheus")
}

type ThanosRuler struct {
	resource *promresourcesv1.ThanosRuler
	informer prominformersv1.PrometheusRuleInformer
	client   promresourcesclient.Interface
J
junotx 已提交
275
	locker   locker.Locker
J
junotx 已提交
276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352
}

func NewThanosRuler(resource *promresourcesv1.ThanosRuler, informer prominformersv1.PrometheusRuleInformer,
	client promresourcesclient.Interface) Ruler {
	return &ThanosRuler{
		resource: resource,
		informer: informer,
		client:   client,
	}
}

func (r *ThanosRuler) Namespace() string {
	return r.resource.Namespace
}

func (r *ThanosRuler) RuleResourceNamespaceSelector() (labels.Selector, error) {
	if r.resource.Spec.RuleNamespaceSelector == nil {
		return nil, nil
	}
	return metav1.LabelSelectorAsSelector(r.resource.Spec.RuleNamespaceSelector)
}

func (r *ThanosRuler) RuleResourceSelector(extraRuleSelector labels.Selector) (labels.Selector, error) {
	rSelector, err := metav1.LabelSelectorAsSelector(r.resource.Spec.RuleSelector)
	if err != nil {
		return nil, err
	}
	if extraRuleSelector != nil {
		if requirements, ok := extraRuleSelector.Requirements(); ok {
			rSelector = rSelector.Add(requirements...)
		}
	}
	return rSelector, nil
}

func (r *ThanosRuler) ExternalLabels() func() map[string]string {
	// rules gotten from thanos ruler endpoint include the labels
	lbls := make(map[string]string)
	if ls := r.resource.Spec.Labels; ls != nil {
		for k, v := range ls {
			lbls[k] = v
		}
	}
	return func() map[string]string {
		return lbls
	}
}

func (r *ThanosRuler) ListRuleResources(ruleNamespace *corev1.Namespace, extraRuleSelector labels.Selector) (
	[]*promresourcesv1.PrometheusRule, error) {
	selected, err := ruleNamespaceSelected(r, ruleNamespace)
	if err != nil {
		return nil, err
	}
	if !selected {
		return nil, nil
	}
	rSelector, err := r.RuleResourceSelector(extraRuleSelector)
	if err != nil {
		return nil, err
	}
	return r.informer.Lister().PrometheusRules(ruleNamespace.Name).List(rSelector)
}

func (r *ThanosRuler) AddAlertingRule(ctx context.Context, ruleNamespace *corev1.Namespace,
	extraRuleResourceSelector labels.Selector,
	group string, rule *promresourcesv1.Rule, ruleResourceLabels map[string]string) error {

	prometheusRules, err := r.ListRuleResources(ruleNamespace, extraRuleResourceSelector)
	if err != nil {
		return err
	}

	return r.addAlertingRule(ctx, ruleNamespace, prometheusRules, nil, group, rule, ruleResourceLabels)
}

func (r *ThanosRuler) addAlertingRule(ctx context.Context, ruleNamespace *corev1.Namespace,
J
junotx 已提交
353
	prometheusRules []*promresourcesv1.PrometheusRule, excludePrometheusRules map[string]*promresourcesv1.PrometheusRule,
J
junotx 已提交
354 355 356 357 358 359 360
	group string, rule *promresourcesv1.Rule, ruleResourceLabels map[string]string) error {

	sort.Slice(prometheusRules, func(i, j int) bool {
		return len(fmt.Sprint(prometheusRules[i])) <= len(fmt.Sprint(prometheusRules[j]))
	})

	for _, prometheusRule := range prometheusRules {
J
junotx 已提交
361 362
		if len(excludePrometheusRules) > 0 {
			if _, ok := excludePrometheusRules[prometheusRule.Name]; ok {
J
junotx 已提交
363 364 365
				continue
			}
		}
J
junotx 已提交
366 367 368 369 370 371 372 373 374 375 376
		if err := r.doRuleResourceOperation(prometheusRule, func(newerPr *promresourcesv1.PrometheusRule) error {
			resource := ruleResource(*newerPr)
			if ok, err := resource.addAlertingRule(group, rule); err != nil {
				return err
			} else if ok {
				if err = resource.commit(ctx, r.client); err != nil {
					return err
				}
			}
			return nil
		}); err != nil {
J
junotx 已提交
377 378
			if err == errOutOfConfigMapSize {
				break
J
junotx 已提交
379 380
			} else if resourceNotFound(err) {
				continue
J
junotx 已提交
381 382 383
			}
			return err
		}
J
junotx 已提交
384
		return nil
J
junotx 已提交
385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417
	}
	// create a new rule resource and add rule into it when all existing rule resources are full.
	newPromRule := promresourcesv1.PrometheusRule{
		ObjectMeta: metav1.ObjectMeta{
			Namespace:    ruleNamespace.Name,
			GenerateName: customAlertingRuleResourcePrefix,
			Labels:       ruleResourceLabels,
		},
		Spec: promresourcesv1.PrometheusRuleSpec{
			Groups: []promresourcesv1.RuleGroup{{
				Name:  group,
				Rules: []promresourcesv1.Rule{*rule},
			}},
		},
	}
	if _, err := r.client.MonitoringV1().
		PrometheusRules(ruleNamespace.Name).Create(ctx, &newPromRule, metav1.CreateOptions{}); err != nil {
		return errors.Wrapf(err, "error creating a prometheusrule resource %s/%s",
			newPromRule.Namespace, newPromRule.Name)
	}
	return nil
}

func (r *ThanosRuler) UpdateAlertingRule(ctx context.Context, ruleNamespace *corev1.Namespace,
	extraRuleResourceSelector labels.Selector,
	group string, rule *promresourcesv1.Rule, ruleResourceLabels map[string]string) error {

	prometheusRules, err := r.ListRuleResources(ruleNamespace, extraRuleResourceSelector)
	if err != nil {
		return err
	}

	var (
J
junotx 已提交
418 419 420
		found        bool
		success      bool
		prsToDelRule = make(map[string]*promresourcesv1.PrometheusRule)
J
junotx 已提交
421
	)
J
junotx 已提交
422
	for i, prometheusRule := range prometheusRules {
J
junotx 已提交
423
		if success { // If the update has been successful, delete the possible same rule in other resources
J
junotx 已提交
424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442
			if err := r.doRuleResourceOperation(prometheusRule, func(newerPr *promresourcesv1.PrometheusRule) error {
				resource := ruleResource(*newerPr)
				if ok, err := resource.deleteAlertingRule(rule.Alert); err != nil {
					return err
				} else if ok {
					if err = resource.commit(ctx, r.client); err != nil {
						return err
					}
				}
				return nil
			}); err != nil && !resourceNotFound(err) {
				return err
			}
			continue
		}

		if err := r.doRuleResourceOperation(prometheusRule, func(newerPr *promresourcesv1.PrometheusRule) error {
			resource := ruleResource(*newerPr)
			if ok, err := resource.updateAlertingRule(group, rule); err != nil {
J
junotx 已提交
443 444 445 446 447 448
				return err
			} else if ok {
				if err = resource.commit(ctx, r.client); err != nil {
					return err
				}
			}
J
junotx 已提交
449 450 451 452 453 454
			return nil
		}); err != nil {
			if resourceNotFound(err) {
				continue
			} else if err == errOutOfConfigMapSize {
				// updating the rule in the resource may oversize the size limit,
J
junotx 已提交
455
				// so delete it and then add the new rule to a new resource.
J
junotx 已提交
456
				prsToDelRule[prometheusRule.Name] = prometheusRules[i]
J
junotx 已提交
457
				found = true
J
junotx 已提交
458
				continue
J
junotx 已提交
459
			}
J
junotx 已提交
460
			return err
J
junotx 已提交
461
		}
J
junotx 已提交
462 463
		found = true
		success = true
J
junotx 已提交
464 465 466
	}

	if !found {
467
		return v2alpha1.ErrAlertingRuleNotFound
J
junotx 已提交
468 469 470
	}

	if !success {
J
junotx 已提交
471
		err := r.addAlertingRule(ctx, ruleNamespace, prometheusRules, prsToDelRule, group, rule, ruleResourceLabels)
J
junotx 已提交
472 473 474 475
		if err != nil {
			return err
		}
	}
J
junotx 已提交
476 477 478 479
	for _, pr := range prsToDelRule {
		if err := r.doRuleResourceOperation(pr, func(newerPr *promresourcesv1.PrometheusRule) error {
			resource := ruleResource(*newerPr)
			if ok, err := resource.deleteAlertingRule(rule.Alert); err != nil {
J
junotx 已提交
480
				return err
J
junotx 已提交
481 482 483 484
			} else if ok {
				if err = resource.commit(ctx, r.client); err != nil {
					return err
				}
J
junotx 已提交
485
			}
J
junotx 已提交
486 487 488
			return nil
		}); err != nil && !resourceNotFound(err) {
			return err
J
junotx 已提交
489 490 491 492 493 494 495 496 497 498 499 500 501
		}
	}
	return nil
}

func (r *ThanosRuler) DeleteAlertingRule(ctx context.Context, ruleNamespace *corev1.Namespace,
	extraRuleResourceSelector labels.Selector, group string, name string) error {
	prometheusRules, err := r.ListRuleResources(ruleNamespace, extraRuleResourceSelector)
	if err != nil {
		return err
	}
	var success bool
	for _, prometheusRule := range prometheusRules {
J
junotx 已提交
502 503 504
		if err := r.doRuleResourceOperation(prometheusRule, func(newerPr *promresourcesv1.PrometheusRule) error {
			resource := ruleResource(*newerPr)
			if ok, err := resource.deleteAlertingRule(name); err != nil {
J
junotx 已提交
505
				return err
J
junotx 已提交
506 507 508 509
			} else if ok {
				if err = resource.commit(ctx, r.client); err != nil {
					return err
				}
J
junotx 已提交
510
			}
J
junotx 已提交
511 512 513 514 515 516
			return nil
		}); err != nil {
			if resourceNotFound(err) {
				continue
			}
			return err
J
junotx 已提交
517
		}
J
junotx 已提交
518
		success = true
J
junotx 已提交
519 520
	}
	if !success {
521
		return v2alpha1.ErrAlertingRuleNotFound
J
junotx 已提交
522 523 524 525
	}
	return nil
}

J
junotx 已提交
526 527 528 529 530 531 532 533 534 535 536 537 538 539
func (r *ThanosRuler) doRuleResourceOperation(pr *promresourcesv1.PrometheusRule,
	operation func(newerPr *promresourcesv1.PrometheusRule) error) error {
	key := pr.Namespace + "/" + pr.Name
	return retry.RetryOnConflict(retry.DefaultRetry, func() error {
		r.locker.Lock(key)
		defer r.locker.Unlock(key)
		pr, err := r.informer.Lister().PrometheusRules(pr.Namespace).Get(pr.Name)
		if err != nil {
			return err
		}
		return operation(pr)
	})
}

J
junotx 已提交
540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555
func ruleNamespaceSelected(r Ruler, ruleNamespace *corev1.Namespace) (bool, error) {
	rnSelector, err := r.RuleResourceNamespaceSelector()
	if err != nil {
		return false, err
	}
	if rnSelector == nil { // refer to the comment of Prometheus.Spec.RuleResourceNamespaceSelector
		if r.Namespace() != ruleNamespace.Name {
			return false, nil
		}
	} else {
		if !rnSelector.Matches(labels.Set(ruleNamespace.Labels)) {
			return false, nil
		}
	}
	return true, nil
}
J
junotx 已提交
556 557 558 559 560 561 562 563 564 565

func resourceNotFound(err error) bool {
	switch e := err.(type) {
	case *apierrors.StatusError:
		if e.Status().Code == http.StatusNotFound {
			return true
		}
	}
	return false
}