import (
"bufio"
+ "encoding/json"
"flag"
"fmt"
- //"git.curoverse.com/arvados.git/sdk/go/keepclient"
"git.curoverse.com/arvados.git/sdk/go/arvadosclient"
"git.curoverse.com/arvados.git/sdk/go/blockdigest"
"git.curoverse.com/arvados.git/sdk/go/logger"
- "git.curoverse.com/arvados.git/sdk/go/manifest"
- "git.curoverse.com/arvados.git/sdk/go/util"
"git.curoverse.com/arvados.git/services/datamanager/loggerutil"
"io/ioutil"
"log"
"strconv"
"strings"
"sync"
+ "time"
)
type ServerAddress struct {
Host string `json:"service_host"`
Port int `json:"service_port"`
+ Uuid string `json:"uuid"`
}
// Info about a particular block returned by the server
type BlockInfo struct {
- Digest blockdigest.BlockDigest
- Size int
- Mtime int // TODO(misha): Replace this with a timestamp.
+ Digest blockdigest.DigestWithSize
+ Mtime int64 // TODO(misha): Replace this with a timestamp.
}
// Info about a specified block given by a server
type BlockServerInfo struct {
ServerIndex int
- Size int
- Mtime int // TODO(misha): Replace this with a timestamp.
+ Mtime int64 // TODO(misha): Replace this with a timestamp.
}
type ServerContents struct {
- BlockDigestToInfo map[blockdigest.BlockDigest]BlockInfo
+ BlockDigestToInfo map[blockdigest.DigestWithSize]BlockInfo
}
type ServerResponse struct {
KeepServerIndexToAddress []ServerAddress
KeepServerAddressToIndex map[ServerAddress]int
ServerToContents map[ServerAddress]ServerContents
- BlockToServers map[blockdigest.BlockDigest][]BlockServerInfo
+ BlockToServers map[blockdigest.DigestWithSize][]BlockServerInfo
BlockReplicationCounts map[int]int
}
KeepServers []ServerAddress `json:"items"`
}
-// Methods to implement util.SdkListResponse Interface
-func (k KeepServiceList) NumItemsAvailable() (numAvailable int, err error) {
- return k.ItemsAvailable, nil
-}
-
-func (k KeepServiceList) NumItemsContained() (numContained int, err error) {
- return len(k.KeepServers), nil
-}
-
var (
// Don't access the token directly, use getDataManagerToken() to
// make sure it's been read.
"File with the API token we should use to contact keep servers.")
}
+// TODO(misha): Change this to include the UUID as well.
func (s ServerAddress) String() string {
+ return s.HostPort()
+}
+
+func (s ServerAddress) HostPort() string {
return fmt.Sprintf("%s:%d", s.Host, s.Port)
}
results = GetKeepServers(params)
log.Printf("Returned %d keep disks", len(results.ServerToContents))
- ComputeBlockReplicationCounts(&results)
+ results.Summarize(params.Logger)
log.Printf("Replication level distribution: %v",
results.BlockReplicationCounts)
"contain a valid ArvadosClient, but instead it is nil.")
}
- sdkParams := arvadosclient.Dict{}
+ sdkParams := arvadosclient.Dict{
+ "filters": [][]string{[]string{"service_type", "=", "disk"}},
+ }
if params.Limit > 0 {
sdkParams["limit"] = params.Limit
}
var sdkResponse KeepServiceList
- err := params.Client.Call("GET", "keep_services", "", "accessible", sdkParams, &sdkResponse)
+ err := params.Client.List("keep_services", sdkParams, &sdkResponse)
if err != nil {
loggerutil.FatalWithMessage(params.Logger,
fmt.Sprintf("Error requesting keep disks from API server: %v", err))
}
- // TODO(misha): Rewrite this block, stop using ContainsAllAvailableItems()
- {
- var numReceived, numAvailable int
- results.ReadAllServers, numReceived, numAvailable =
- util.ContainsAllAvailableItems(sdkResponse)
-
- if !results.ReadAllServers {
- log.Printf("ERROR: Did not receive all keep server addresses.")
- }
- log.Printf("Received %d of %d available keep server addresses.",
- numReceived,
- numAvailable)
- }
-
if params.Logger != nil {
- properties, _ := params.Logger.Edit()
- keepInfo := make(map[string]interface{})
-
- keepInfo["num_keep_servers_available"] = sdkResponse.ItemsAvailable
- keepInfo["num_keep_servers_received"] = len(sdkResponse.KeepServers)
- keepInfo["keep_servers"] = sdkResponse.KeepServers
+ params.Logger.Update(func(p map[string]interface{}, e map[string]interface{}) {
+ keepInfo := logger.GetOrCreateMap(p, "keep_info")
+ keepInfo["num_keep_servers_available"] = sdkResponse.ItemsAvailable
+ keepInfo["num_keep_servers_received"] = len(sdkResponse.KeepServers)
+ keepInfo["keep_servers"] = sdkResponse.KeepServers
+ })
+ }
- properties["keep_info"] = keepInfo
+ log.Printf("Received keep services list: %+v", sdkResponse)
- params.Logger.Record()
+ if len(sdkResponse.KeepServers) < sdkResponse.ItemsAvailable {
+ loggerutil.FatalWithMessage(params.Logger,
+ fmt.Sprintf("Did not receive all available keep servers: %+v", sdkResponse))
}
- log.Printf("Received keep services list: %v", sdkResponse)
-
results.KeepServerIndexToAddress = sdkResponse.KeepServers
results.KeepServerAddressToIndex = make(map[ServerAddress]int)
for i, address := range results.KeepServerIndexToAddress {
}
results.ServerToContents = make(map[ServerAddress]ServerContents)
- results.BlockToServers = make(map[blockdigest.BlockDigest][]BlockServerInfo)
+ results.BlockToServers = make(map[blockdigest.DigestWithSize][]BlockServerInfo)
// Read all the responses
for i := range sdkResponse.KeepServers {
results.BlockToServers[blockInfo.Digest] = append(
results.BlockToServers[blockInfo.Digest],
BlockServerInfo{ServerIndex: serverIndex,
- Size: blockInfo.Size,
Mtime: blockInfo.Mtime})
}
}
keepServer ServerAddress,
client http.Client) (response ServerResponse) {
+ GetServerStatus(arvLogger, keepServer, client)
+
req := CreateIndexRequest(arvLogger, keepServer)
resp, err := client.Do(req)
if err != nil {
- log.Fatalf("Error fetching %s: %v", req.URL.String(), err)
+ loggerutil.FatalWithMessage(arvLogger,
+ fmt.Sprintf("Error fetching %s: %v. Response was %+v",
+ req.URL.String(),
+ err,
+ resp))
}
return ReadServerResponse(arvLogger, keepServer, resp)
}
+func GetServerStatus(arvLogger *logger.Logger,
+ keepServer ServerAddress,
+ client http.Client) {
+ url := fmt.Sprintf("http://%s:%d/status.json",
+ keepServer.Host,
+ keepServer.Port)
+
+ if arvLogger != nil {
+ now := time.Now()
+ arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
+ keepInfo := logger.GetOrCreateMap(p, "keep_info")
+ serverInfo := make(map[string]interface{})
+ serverInfo["status_request_sent_at"] = now
+ serverInfo["host"] = keepServer.Host
+ serverInfo["port"] = keepServer.Port
+
+ keepInfo[keepServer.Uuid] = serverInfo
+ })
+ }
+
+ resp, err := client.Get(url)
+ if err != nil {
+ loggerutil.FatalWithMessage(arvLogger,
+ fmt.Sprintf("Error getting keep status from %s: %v", url, err))
+ } else if resp.StatusCode != 200 {
+ loggerutil.FatalWithMessage(arvLogger,
+ fmt.Sprintf("Received error code %d in response to request "+
+ "for %s status: %s",
+ resp.StatusCode, url, resp.Status))
+ }
+
+ var keepStatus map[string]interface{}
+ decoder := json.NewDecoder(resp.Body)
+ decoder.UseNumber()
+ err = decoder.Decode(&keepStatus)
+ if err != nil {
+ loggerutil.FatalWithMessage(arvLogger,
+ fmt.Sprintf("Error decoding keep status from %s: %v", url, err))
+ }
+
+ if arvLogger != nil {
+ now := time.Now()
+ arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
+ keepInfo := logger.GetOrCreateMap(p, "keep_info")
+ serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{})
+ serverInfo["status_response_processed_at"] = now
+ serverInfo["status"] = keepStatus
+ })
+ }
+}
+
func CreateIndexRequest(arvLogger *logger.Logger,
keepServer ServerAddress) (req *http.Request) {
- // Create and send request.
url := fmt.Sprintf("http://%s:%d/index", keepServer.Host, keepServer.Port)
log.Println("About to fetch keep server contents from " + url)
+ if arvLogger != nil {
+ now := time.Now()
+ arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
+ keepInfo := logger.GetOrCreateMap(p, "keep_info")
+ serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{})
+ serverInfo["index_request_sent_at"] = now
+ })
+ }
+
req, err := http.NewRequest("GET", url, nil)
if err != nil {
- log.Fatalf("Error building http request for %s: %v", url, err)
+ loggerutil.FatalWithMessage(arvLogger,
+ fmt.Sprintf("Error building http request for %s: %v", url, err))
}
req.Header.Add("Authorization",
resp *http.Response) (response ServerResponse) {
if resp.StatusCode != 200 {
- log.Fatalf("Received error code %d in response to request for %s index: %s",
- resp.StatusCode, keepServer.String(), resp.Status)
+ loggerutil.FatalWithMessage(arvLogger,
+ fmt.Sprintf("Received error code %d in response to request "+
+ "for %s index: %s",
+ resp.StatusCode, keepServer.String(), resp.Status))
+ }
+
+ if arvLogger != nil {
+ now := time.Now()
+ arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
+ keepInfo := logger.GetOrCreateMap(p, "keep_info")
+ serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{})
+ serverInfo["index_response_received_at"] = now
+ })
}
response.Address = keepServer
response.Contents.BlockDigestToInfo =
- make(map[blockdigest.BlockDigest]BlockInfo)
+ make(map[blockdigest.DigestWithSize]BlockInfo)
scanner := bufio.NewScanner(resp.Body)
numLines, numDuplicates, numSizeDisagreements := 0, 0, 0
for scanner.Scan() {
numLines++
blockInfo, err := parseBlockInfoFromIndexLine(scanner.Text())
if err != nil {
- log.Fatalf("Error parsing BlockInfo from index line received from %s: %v",
- keepServer.String(),
- err)
+ loggerutil.FatalWithMessage(arvLogger,
+ fmt.Sprintf("Error parsing BlockInfo from index line "+
+ "received from %s: %v",
+ keepServer.String(),
+ err))
}
if storedBlock, ok := response.Contents.BlockDigestToInfo[blockInfo.Digest]; ok {
// This server returned multiple lines containing the same block digest.
numDuplicates += 1
- if storedBlock.Size != blockInfo.Size {
- numSizeDisagreements += 1
- // TODO(misha): Consider failing here.
- log.Printf("Saw different sizes for the same block on %s: %v %v",
- keepServer.String(),
- storedBlock,
- blockInfo)
- }
- // Keep the block that is bigger, or the block that's newer in
- // the case of a size tie.
- if storedBlock.Size < blockInfo.Size ||
- (storedBlock.Size == blockInfo.Size &&
- storedBlock.Mtime < blockInfo.Mtime) {
+ // Keep the block that's newer.
+ if storedBlock.Mtime < blockInfo.Mtime {
response.Contents.BlockDigestToInfo[blockInfo.Digest] = blockInfo
}
} else {
}
}
if err := scanner.Err(); err != nil {
- log.Fatalf("Received error scanning index response from %s: %v",
- keepServer.String(),
- err)
+ loggerutil.FatalWithMessage(arvLogger,
+ fmt.Sprintf("Received error scanning index response from %s: %v",
+ keepServer.String(),
+ err))
} else {
log.Printf("%s index contained %d lines with %d duplicates with "+
"%d size disagreements",
numLines,
numDuplicates,
numSizeDisagreements)
+
+ if arvLogger != nil {
+ now := time.Now()
+ arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
+ keepInfo := logger.GetOrCreateMap(p, "keep_info")
+ serverInfo := keepInfo[keepServer.Uuid].(map[string]interface{})
+
+ serverInfo["processing_finished_at"] = now
+ serverInfo["lines_received"] = numLines
+ serverInfo["duplicates_seen"] = numDuplicates
+ serverInfo["size_disagreements_seen"] = numSizeDisagreements
+ })
+ }
}
resp.Body.Close()
return
tokens)
}
- var locator manifest.BlockLocator
- if locator, err = manifest.ParseBlockLocator(tokens[0]); err != nil {
+ var locator blockdigest.BlockLocator
+ if locator, err = blockdigest.ParseBlockLocator(tokens[0]); err != nil {
+ err = fmt.Errorf("%v Received error while parsing line \"%s\"",
+ err, indexLine)
return
}
if len(locator.Hints) > 0 {
return
}
- blockInfo.Mtime, err = strconv.Atoi(tokens[1])
+ blockInfo.Mtime, err = strconv.ParseInt(tokens[1], 10, 64)
if err != nil {
return
}
- blockInfo.Digest = locator.Digest
- blockInfo.Size = locator.Size
+ blockInfo.Digest =
+ blockdigest.DigestWithSize{Digest: locator.Digest,
+ Size: uint32(locator.Size)}
return
}
-func ComputeBlockReplicationCounts(readServers *ReadServers) {
+func (readServers *ReadServers) Summarize(arvLogger *logger.Logger) {
readServers.BlockReplicationCounts = make(map[int]int)
for _, infos := range readServers.BlockToServers {
replication := len(infos)
readServers.BlockReplicationCounts[replication] += 1
}
+
+ if arvLogger != nil {
+ arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) {
+ keepInfo := logger.GetOrCreateMap(p, "keep_info")
+ keepInfo["distinct_blocks_stored"] = len(readServers.BlockToServers)
+ })
+ }
+
}