crowdsec/pkg/database/flush.go
mmetc f8f0b2a211
improve support for parsing time durations with 'day' units (#3599)
* custom duration type for "cscli decisions list", "cscli alerts list"
* custom duration type for "cscli allowlist add"
* custom duration type for "cscli machines prune"
* custom duration type for "cscli bouncers prune"
* replace old function ParseDuration
* use custom duration type in expr helpers
* update dependency
* lint
* test fix
* support days in 'metrics_max_age'
* DurationWithDays for 'max_age'
2025-05-05 15:12:29 +02:00

328 lines
9.5 KiB
Go

package database
import (
"context"
"errors"
"fmt"
"time"
"github.com/go-co-op/gocron"
log "github.com/sirupsen/logrus"
"github.com/crowdsecurity/go-cs-lib/cstime"
"github.com/crowdsecurity/crowdsec/pkg/csconfig"
"github.com/crowdsecurity/crowdsec/pkg/database/ent/alert"
"github.com/crowdsecurity/crowdsec/pkg/database/ent/allowlistitem"
"github.com/crowdsecurity/crowdsec/pkg/database/ent/bouncer"
"github.com/crowdsecurity/crowdsec/pkg/database/ent/decision"
"github.com/crowdsecurity/crowdsec/pkg/database/ent/event"
"github.com/crowdsecurity/crowdsec/pkg/database/ent/machine"
"github.com/crowdsecurity/crowdsec/pkg/database/ent/metric"
"github.com/crowdsecurity/crowdsec/pkg/types"
)
const (
// how long to keep metrics in the local database
defaultMetricsMaxAge = 7 * 24 * time.Hour
flushInterval = 1 * time.Minute
)
func (c *Client) StartFlushScheduler(ctx context.Context, config *csconfig.FlushDBCfg) (*gocron.Scheduler, error) {
maxItems := 0
if config.MaxItems != nil && *config.MaxItems <= 0 {
return nil, errors.New("max_items can't be zero or negative")
}
if config.MaxItems != nil {
maxItems = *config.MaxItems
}
// Init & Start cronjob every minute for alerts
scheduler := gocron.NewScheduler(time.UTC)
job, err := scheduler.Every(1).Minute().Do(c.FlushAlerts, ctx, time.Duration(config.MaxAge), maxItems)
if err != nil {
return nil, fmt.Errorf("while starting FlushAlerts scheduler: %w", err)
}
job.SingletonMode()
// Init & Start cronjob every hour for bouncers/agents
if config.AgentsGC != nil {
if config.AgentsGC.Cert != nil {
duration, err := cstime.ParseDurationWithDays(*config.AgentsGC.Cert)
if err != nil {
return nil, fmt.Errorf("while parsing agents cert auto-delete duration: %w", err)
}
config.AgentsGC.CertDuration = &duration
}
if config.AgentsGC.LoginPassword != nil {
duration, err := cstime.ParseDurationWithDays(*config.AgentsGC.LoginPassword)
if err != nil {
return nil, fmt.Errorf("while parsing agents login/password auto-delete duration: %w", err)
}
config.AgentsGC.LoginPasswordDuration = &duration
}
if config.AgentsGC.Api != nil {
log.Warning("agents auto-delete for API auth is not supported (use cert or login_password)")
}
}
if config.BouncersGC != nil {
if config.BouncersGC.Cert != nil {
duration, err := cstime.ParseDurationWithDays(*config.BouncersGC.Cert)
if err != nil {
return nil, fmt.Errorf("while parsing bouncers cert auto-delete duration: %w", err)
}
config.BouncersGC.CertDuration = &duration
}
if config.BouncersGC.Api != nil {
duration, err := cstime.ParseDurationWithDays(*config.BouncersGC.Api)
if err != nil {
return nil, fmt.Errorf("while parsing bouncers api auto-delete duration: %w", err)
}
config.BouncersGC.ApiDuration = &duration
}
if config.BouncersGC.LoginPassword != nil {
log.Warning("bouncers auto-delete for login/password auth is not supported (use cert or api)")
}
}
baJob, err := scheduler.Every(flushInterval).Do(c.FlushAgentsAndBouncers, ctx, config.AgentsGC, config.BouncersGC)
if err != nil {
return nil, fmt.Errorf("while starting FlushAgentsAndBouncers scheduler: %w", err)
}
baJob.SingletonMode()
metricsJob, err := scheduler.Every(flushInterval).Do(c.flushMetrics, ctx, time.Duration(config.MetricsMaxAge))
if err != nil {
return nil, fmt.Errorf("while starting flushMetrics scheduler: %w", err)
}
metricsJob.SingletonMode()
allowlistsJob, err := scheduler.Every(flushInterval).Do(c.flushAllowlists, ctx)
if err != nil {
return nil, fmt.Errorf("while starting FlushAllowlists scheduler: %w", err)
}
allowlistsJob.SingletonMode()
scheduler.StartAsync()
return scheduler, nil
}
// flushMetrics deletes metrics older than maxAge, regardless if they have been pushed to CAPI or not
func (c *Client) flushMetrics(ctx context.Context, maxAge time.Duration) {
if maxAge == 0 {
maxAge = defaultMetricsMaxAge
}
c.Log.Debugf("flushing metrics older than %s", maxAge)
deleted, err := c.Ent.Metric.Delete().Where(
metric.ReceivedAtLTE(time.Now().UTC().Add(-maxAge)),
).Exec(ctx)
if err != nil {
c.Log.Errorf("while flushing metrics: %s", err)
return
}
if deleted > 0 {
c.Log.Debugf("flushed %d metrics snapshots", deleted)
}
}
func (c *Client) FlushOrphans(ctx context.Context) {
/* While it has only been linked to some very corner-case bug : https://github.com/crowdsecurity/crowdsec/issues/778 */
/* We want to take care of orphaned events for which the parent alert/decision has been deleted */
eventsCount, err := c.Ent.Event.Delete().Where(event.Not(event.HasOwner())).Exec(ctx)
if err != nil {
c.Log.Warningf("error while deleting orphan events: %s", err)
return
}
if eventsCount > 0 {
c.Log.Infof("%d deleted orphan events", eventsCount)
}
eventsCount, err = c.Ent.Decision.Delete().Where(
decision.Not(decision.HasOwner())).Where(decision.UntilLTE(time.Now().UTC())).Exec(ctx)
if err != nil {
c.Log.Warningf("error while deleting orphan decisions: %s", err)
return
}
if eventsCount > 0 {
c.Log.Infof("%d deleted orphan decisions", eventsCount)
}
}
func (c *Client) flushBouncers(ctx context.Context, authType string, duration *time.Duration) {
if duration == nil {
return
}
count, err := c.Ent.Bouncer.Delete().Where(
bouncer.LastPullLTE(time.Now().UTC().Add(-*duration)),
).Where(
bouncer.AuthTypeEQ(authType),
).Exec(ctx)
if err != nil {
c.Log.Errorf("while auto-deleting expired bouncers (%s): %s", authType, err)
return
}
if count > 0 {
c.Log.Infof("deleted %d expired bouncers (%s)", count, authType)
}
}
func (c *Client) flushAgents(ctx context.Context, authType string, duration *time.Duration) {
if duration == nil {
return
}
count, err := c.Ent.Machine.Delete().Where(
machine.LastHeartbeatLTE(time.Now().UTC().Add(-*duration)),
machine.Not(machine.HasAlerts()),
machine.AuthTypeEQ(authType),
).Exec(ctx)
if err != nil {
c.Log.Errorf("while auto-deleting expired machines (%s): %s", authType, err)
return
}
if count > 0 {
c.Log.Infof("deleted %d expired machines (%s auth)", count, authType)
}
}
func (c *Client) FlushAgentsAndBouncers(ctx context.Context, agentsCfg *csconfig.AuthGCCfg, bouncersCfg *csconfig.AuthGCCfg) error {
log.Debug("starting FlushAgentsAndBouncers")
if agentsCfg != nil {
c.flushAgents(ctx, types.TlsAuthType, agentsCfg.CertDuration)
c.flushAgents(ctx, types.PasswordAuthType, agentsCfg.LoginPasswordDuration)
}
if bouncersCfg != nil {
c.flushBouncers(ctx, types.TlsAuthType, bouncersCfg.CertDuration)
c.flushBouncers(ctx, types.ApiKeyAuthType, bouncersCfg.ApiDuration)
}
return nil
}
func (c *Client) FlushAlerts(ctx context.Context, maxAge time.Duration, maxItems int) error {
var (
deletedByAge int
deletedByNbItem int
totalAlerts int
err error
)
if !c.CanFlush {
c.Log.Debug("a list is being imported, flushing later")
return nil
}
c.Log.Debug("Flushing orphan alerts")
c.FlushOrphans(ctx)
c.Log.Debug("Done flushing orphan alerts")
totalAlerts, err = c.TotalAlerts(ctx)
if err != nil {
c.Log.Warningf("FlushAlerts (max items count): %s", err)
return fmt.Errorf("unable to get alerts count: %w", err)
}
c.Log.Debugf("FlushAlerts (Total alerts): %d", totalAlerts)
if maxAge != 0 {
filter := map[string][]string{
"created_before": {maxAge.String()},
}
nbDeleted, err := c.DeleteAlertWithFilter(ctx, filter)
if err != nil {
c.Log.Warningf("FlushAlerts (max age): %s", err)
return fmt.Errorf("unable to flush alerts with filter until=%s: %w", maxAge, err)
}
c.Log.Debugf("FlushAlerts (deleted max age alerts): %d", nbDeleted)
deletedByAge = nbDeleted
}
if maxItems > 0 {
// We get the highest id for the alerts
// We subtract MaxItems to avoid deleting alerts that are not old enough
// This gives us the oldest alert that we want to keep
// We then delete all the alerts with an id lower than this one
// We can do this because the id is auto-increment, and the database won't reuse the same id twice
lastAlert, err := c.QueryAlertWithFilter(ctx, map[string][]string{
"sort": {"DESC"},
"limit": {"1"},
// we do not care about fetching the edges, we just want the id
"with_decisions": {"false"},
})
c.Log.Debugf("FlushAlerts (last alert): %+v", lastAlert)
if err != nil {
c.Log.Errorf("FlushAlerts: could not get last alert: %s", err)
return fmt.Errorf("could not get last alert: %w", err)
}
if len(lastAlert) != 0 {
maxid := lastAlert[0].ID - maxItems
c.Log.Debugf("FlushAlerts (max id): %d", maxid)
if maxid > 0 {
// This may lead to orphan alerts (at least on MySQL), but the next time the flush job will run, they will be deleted
deletedByNbItem, err = c.Ent.Alert.Delete().Where(alert.IDLT(maxid)).Exec(ctx)
if err != nil {
c.Log.Errorf("FlushAlerts: Could not delete alerts: %s", err)
return fmt.Errorf("could not delete alerts: %w", err)
}
}
}
}
if deletedByNbItem > 0 {
c.Log.Infof("flushed %d/%d alerts because the max number of alerts has been reached (%d max)",
deletedByNbItem, totalAlerts, maxItems)
}
if deletedByAge > 0 {
c.Log.Infof("flushed %d/%d alerts because they were created %s ago or more",
deletedByAge, totalAlerts, maxAge)
}
return nil
}
func (c *Client) flushAllowlists(ctx context.Context) {
deleted, err := c.Ent.AllowListItem.Delete().Where(
allowlistitem.ExpiresAtLTE(time.Now().UTC()),
).Exec(ctx)
if err != nil {
c.Log.Errorf("while flushing allowlists: %s", err)
return
}
if deleted > 0 {
c.Log.Debugf("flushed %d allowlists", deleted)
}
}