covidnetes/woodpecker-autoscaler
4
0
Fork 0
Code Issues 1 Pull Requests Packages Projects Releases 3 Wiki Activity

(feat) Cost optimized mode
All checks were successful
ci/woodpecker/pr/pr Pipeline was successful
Details

Browse Source
This commit is contained in:
Tobias Trabelsi
2023-12-31 23:02:32 +01:00
parent a190c1e0fe
commit 30a219c91f
5 changed files with 135 additions and 68 deletions
Download Patch File Download Diff File Expand all files Collapse all files

170
cmd/woodpecker-autoscaler.go
View File

@ -13,6 +13,106 @@ import (
log "github.com/sirupsen/logrus"
)
func SpawnNewAgent(cfg *config.Config) {
agent, err := woodpecker.CreateWoodpeckerAgent(cfg)
if err != nil {
log.WithFields(log.Fields{
"Caller": "SpawnNewAgent",
}).Fatal(fmt.Sprintf("Error creating new agent: %s", err.Error()))
}
server, err := hetzner.CreateNewAgent(cfg, agent)
if err != nil {
log.WithFields(log.Fields{
"Caller": "SpawnNewAgent",
}).Fatal(fmt.Sprintf("Error spawning new agent: %s", err.Error()))
}
for {
server, err = hetzner.RefreshNodeInfo(cfg, server.ID)
if err != nil {
log.WithFields(log.Fields{
"Caller": "SpawnNewAgent",
}).Fatal(fmt.Sprintf("Failed to start Agent: %s", err.Error()))
}
if server.Status == hcloud.ServerStatusRunning {
log.WithFields(log.Fields{
"Caller": "SpawnNewAgent",
}).Infof("%s started!", server.Name)
break
}
log.WithFields(log.Fields{
"Caller": "SpawnNewAgent",
}).Infof("%s is in status %s", server.Name, server.Status)
time.Sleep(30 * time.Second)
}
}
func CheckJobs(cfg *config.Config, ownedNodes []hcloud.Server, pendingTasks int) {
log.WithFields(log.Fields{
"Caller": "CheckJobs",
}).Info("Checking if agents can be removed")
runningTasks, err := woodpecker.CheckRunning(cfg)
if err != nil {
log.WithFields(log.Fields{
"Caller": "CheckJobs",
}).Fatal(fmt.Sprintf("Error checking woodpecker queue: %s", err.Error()))
}
if (runningTasks <= len(ownedNodes) && runningTasks != 0) || pendingTasks > 0 {
log.WithFields(log.Fields{
"Caller": "CheckJobs",
}).Info("Still found running tasks. No agent to be removed")
} else {
if len(ownedNodes) == 0 {
log.WithFields(log.Fields{
"Caller": "CheckJobs",
}).Info("Nothing running and not owning any nodes")
} else {
log.WithFields(log.Fields{
"Caller": "CheckJobs",
}).Info("No tasks running. Will remove agents")
Decom(cfg, ownedNodes)
}
}
}
func Decom(cfg *config.Config, ownedNodes []hcloud.Server) {
for _, server := range ownedNodes {
if cfg.CostOptimizedMode {
runtime, err := hetzner.CheckRuntime(cfg, &server)
if err != nil {
log.WithFields(log.Fields{
"Caller": "Decom",
}).Warnf("Error while checking runtime of node %s: %s", server.Name, err.Error())
}
log.WithFields(log.Fields{
"Caller": "Decom",
}).Debugf("Node %s is running for %f", server.Name, runtime.Minutes())
// Check if next check if sooner than the 60 Minute mark of the next hetzner check
// https://docs.hetzner.com/cloud/billing/faq/#how-do-you-bill-your-servers
if (runtime + time.Duration(cfg.CheckInterval)*time.Minute) < 60 {
log.WithFields(log.Fields{
"Caller": "Decom",
}).Infof("Skipping node termination of %s (running for %f Minutes) in Cost Optimized Mode", server.Name, runtime.Minutes())
continue
}
}
agentId, err := hetzner.DecomNode(cfg, &server)
if err != nil {
log.WithFields(log.Fields{
"Caller": "Decom",
}).Warnf("Error while deleting node %s: %s", server.Name, err.Error())
}
err = woodpecker.DecomAgent(cfg, agentId)
if err != nil {
log.WithFields(log.Fields{
"Caller": "Decom",
}).Warnf("Could not delete node %s in woodpecker: %s", server.Name, err.Error())
}
log.WithFields(log.Fields{
"Caller": "Decom",
}).Infof("Deleted node %s", server.Name)
}
}
func main() {
cfg, err := config.GenConfig()
@ -52,75 +152,9 @@ func main() {
"Caller": "Main",
}).Infof("Currently owning %d Agents", len(ownedNodes))
if pendingTasks > len(ownedNodes) {
agent, err := woodpecker.CreateWoodpeckerAgent(cfg)
if err != nil {
log.WithFields(log.Fields{
"Caller": "Main",
}).Fatal(fmt.Sprintf("Error creating new agent: %s", err.Error()))
}
server, err := hetzner.CreateNewAgent(cfg, agent)
if err != nil {
log.WithFields(log.Fields{
"Caller": "Main",
}).Fatal(fmt.Sprintf("Error spawning new agent: %s", err.Error()))
}
for {
server, err = hetzner.RefreshNodeInfo(cfg, server.ID)
if err != nil {
log.WithFields(log.Fields{
"Caller": "Main",
}).Fatal(fmt.Sprintf("Failed to start Agent: %s", err.Error()))
}
if server.Status == hcloud.ServerStatusRunning {
log.WithFields(log.Fields{
"Caller": "Main",
}).Infof("%s started!", server.Name)
break
}
log.WithFields(log.Fields{
"Caller": "Main",
}).Infof("%s is in status %s", server.Name, server.Status)
time.Sleep(30 * time.Second)
}
SpawnNewAgent(cfg)
} else {
log.WithFields(log.Fields{
"Caller": "Main",
}).Info("Checking if agents can be removed")
runningTasks, err := woodpecker.CheckRunning(cfg)
if err != nil {
log.WithFields(log.Fields{
"Caller": "Main",
}).Fatal(fmt.Sprintf("Error checking woodpecker queue: %s", err.Error()))
}
if (runningTasks <= len(ownedNodes) && runningTasks != 0) || pendingTasks > 0 {
log.WithFields(log.Fields{
"Caller": "Main",
}).Info("Still found running tasks. No agent to be removed")
} else {
if len(ownedNodes) == 0 {
log.WithFields(log.Fields{
"Caller": "Main",
}).Info("Nothing running and not owning any nodes")
} else {
log.WithFields(log.Fields{
"Caller": "Main",
}).Info("No tasks running. Will remove agents")
for _, server := range ownedNodes {
agentId, err := hetzner.DecomNode(cfg, &server)
if err != nil {
log.WithFields(log.Fields{
"Caller": "Main",
}).Warnf("Error while deleting node %s: %s", server.Name, err.Error())
}
err = woodpecker.DecomAgent(cfg, agentId)
if err != nil {
log.WithFields(log.Fields{
"Caller": "Main",
}).Warnf("Could not delete node %s in woodpecker: %s", server.Name, err.Error())
}
}
}
}
CheckJobs(cfg, ownedNodes, pendingTasks)
}
log.WithFields(log.Fields{
"Caller": "Main",