mudhorn/prometheus
1
0
Fork 0
mirror of https://github.com/prometheus/prometheus.git synced 2024-11-09 23:24:05 -08:00
Code Issues Actions 3 Packages Projects Releases Wiki Activity

Describe EC2 availability zones at most once per SD load

Browse source 
Closes #9142.

Signed-off-by: George Brighton <george@gebn.co.uk>
This commit is contained in:
George Brighton 2021-08-02 20:52:37 +01:00
parent a715dd84ec
commit 2d8f9c6f61
No known key found for this signature in database
GPG key ID: EB1FF129E25E4DA5
1 changed files with 17 additions and 29 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

46
discovery/aws/ec2.go
View file

@ -194,29 +194,10 @@ func (d *EC2Discovery) ec2Client(ctx context.Context) (*ec2.EC2, error) {
return d.ec2, nil
}
func (d *EC2Discovery) azID(ctx context.Context, az string, azIDsRefreshAttempted *bool) (string, error) {
if azID, ok := d.azToAZID[az]; ok {
return azID, nil
}
if !*azIDsRefreshAttempted {
*azIDsRefreshAttempted = true
if err := d.refreshAZIDs(ctx); err != nil {
return "", err
}
if azID, ok := d.azToAZID[az]; ok {
return azID, nil
}
}
return "", fmt.Errorf("no availability zone ID mapping found for %s", az)
}
func (d *EC2Discovery) refreshAZIDs(ctx context.Context) error {
azs, err := d.ec2.DescribeAvailabilityZonesWithContext(ctx, &ec2.DescribeAvailabilityZonesInput{})
if err != nil {
// Likely due to missing ec2:DescribeAvailabilityZones.
return errors.Wrap(err, "could not describe availability zones")
return err
}
d.azToAZID = make(map[string]string, len(azs.AvailabilityZones))
for _, az := range azs.AvailabilityZones {
@ -243,21 +224,23 @@ func (d *EC2Discovery) refresh(ctx context.Context) ([]*targetgroup.Group, error
})
}
// Only refresh the AZ ID map if we have never been able to build one.
// Prometheus requires a reload if AWS adds a new AZ to the region.
if d.azToAZID == nil {
if err := d.refreshAZIDs(ctx); err != nil {
level.Warn(d.logger).Log(
"msg", "Unable to describe availability zones, ensure Prometheus has ec2:DescribeAvailabilityZones",
"err", err)
}
}
input := &ec2.DescribeInstancesInput{Filters: filters}
azIDsRefreshAttempted := false
if err := ec2Client.DescribeInstancesPagesWithContext(ctx, input, func(p *ec2.DescribeInstancesOutput, lastPage bool) bool {
for _, r := range p.Reservations {
for _, inst := range r.Instances {
if inst.PrivateIpAddress == nil {
continue
}
azID, err := d.azID(ctx, *inst.Placement.AvailabilityZone, &azIDsRefreshAttempted)
if err != nil {
level.Warn(d.logger).Log(
"msg", "Unable to determine availability zone ID",
"az", *inst.Placement.AvailabilityZone,
"err", err)
}
labels := model.LabelSet{
ec2LabelInstanceID: model.LabelValue(*inst.InstanceId),
@ -282,9 +265,14 @@ func (d *EC2Discovery) refresh(ctx context.Context) ([]*targetgroup.Group, error
labels[ec2LabelPublicIP] = model.LabelValue(*inst.PublicIpAddress)
labels[ec2LabelPublicDNS] = model.LabelValue(*inst.PublicDnsName)
}
labels[ec2LabelAMI] = model.LabelValue(*inst.ImageId)
labels[ec2LabelAZ] = model.LabelValue(*inst.Placement.AvailabilityZone)
azID, ok := d.azToAZID[*inst.Placement.AvailabilityZone]
if !ok && d.azToAZID != nil {
level.Debug(d.logger).Log(
"msg", "Availability zone not found, try reloading Prometheus",
"az", *inst.Placement.AvailabilityZone)
}
labels[ec2LabelAZID] = model.LabelValue(azID)
labels[ec2LabelInstanceState] = model.LabelValue(*inst.State.Name)
labels[ec2LabelInstanceType] = model.LabelValue(*inst.InstanceType)