13399: If slurm refuses to renice past 10K, stop trying.

[arvados.git] / services / keep-balance / main.go

diff --git a/services/keep-balance/main.go b/services/keep-balance/main.go
index 04d3e9992b21f74289919dfeb5a67bda509e94f6..90235cbf3188d91bc274412ddd5522dc639fa812 100644 (file)
--- a/services/keep-balance/main.go
+++ b/services/keep-balance/main.go
@@ -1,9 +1,15 @@
+// Copyright (C) The Arvados Authors. All rights reserved.
+//
+// SPDX-License-Identifier: AGPL-3.0
+
 package main
 
 import (
        "encoding/json"
        "flag"
+       "fmt"
        "log"
+       "net/http"
        "os"
        "os/signal"
        "syscall"
@@ -13,6 +19,8 @@ import (
        "git.curoverse.com/arvados.git/sdk/go/config"
 )
 
+var version = "dev"
+
 const defaultConfigPath = "/etc/arvados/keep-balance/keep-balance.yml"
 
 // Config specifies site configuration, like API credentials and the
@@ -38,6 +46,9 @@ type Config struct {
        // more memory, but can reduce store-and-forward latency when
        // fetching pages)
        CollectionBuffers int
+
+       // Timeout for outgoing http request/response cycle.
+       RequestTimeout arvados.Duration
 }
 
 // RunOptions controls runtime behavior. The flags/options that belong
@@ -81,9 +92,16 @@ func main() {
        dumpConfig := flag.Bool("dump-config", false, "write current configuration to stdout and exit")
        dumpFlag := flag.Bool("dump", false, "dump details for each block to stdout")
        debugFlag := flag.Bool("debug", false, "enable debug messages")
+       getVersion := flag.Bool("version", false, "Print version information and exit.")
        flag.Usage = usage
        flag.Parse()
 
+       // Print version information if requested
+       if *getVersion {
+               fmt.Printf("keep-balance %s\n", version)
+               return
+       }
+
        mustReadConfig(&cfg, *configPath)
        if *serviceListPath != "" {
                mustReadConfig(&cfg.KeepServiceList, *serviceListPath)
@@ -93,6 +111,16 @@ func main() {
                log.Fatal(config.DumpAndExit(cfg))
        }
 
+       to := time.Duration(cfg.RequestTimeout)
+       if to == 0 {
+               to = 30 * time.Minute
+       }
+       arvados.DefaultSecureClient.Timeout = to
+       arvados.InsecureHTTPClient.Timeout = to
+       http.DefaultClient.Timeout = to
+
+       log.Printf("keep-balance %s started", version)
+
        if *debugFlag {
                debugf = log.Printf
                if j, err := json.Marshal(cfg); err != nil {