open-nomad/command/agent/alloc_endpoint.go

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

701 lines
19 KiB
Go
Raw Normal View History

2015-09-06 22:37:21 +00:00
package agent
import (
"context"
"encoding/json"
"fmt"
"io"
"net"
2015-09-06 22:37:21 +00:00
"net/http"
"strconv"
2015-09-06 22:37:21 +00:00
"strings"
2016-11-29 00:05:56 +00:00
"github.com/golang/snappy"
"github.com/gorilla/websocket"
"github.com/hashicorp/go-msgpack/codec"
2018-02-06 18:53:00 +00:00
cstructs "github.com/hashicorp/nomad/client/structs"
2015-09-06 22:37:21 +00:00
"github.com/hashicorp/nomad/nomad/structs"
"github.com/hashicorp/nomad/plugins/drivers"
2015-09-06 22:37:21 +00:00
)
const (
allocNotFoundErr = "allocation not found"
resourceNotFoundErr = "resource not found"
)
2015-09-06 22:37:21 +00:00
func (s *HTTPServer) AllocsRequest(resp http.ResponseWriter, req *http.Request) (interface{}, error) {
if req.Method != "GET" {
return nil, CodedError(405, ErrInvalidMethod)
}
args := structs.AllocListRequest{}
if s.parse(resp, req, &args.Region, &args.QueryOptions) {
return nil, nil
}
// Parse resources and task_states field selection
2020-10-14 19:23:25 +00:00
resources, err := parseBool(req, "resources")
if err != nil {
return nil, err
}
2020-10-14 19:23:25 +00:00
taskStates, err := parseBool(req, "task_states")
if err != nil {
return nil, err
}
2020-10-14 19:23:25 +00:00
if resources != nil || taskStates != nil {
args.Fields = structs.NewAllocStubFields()
if resources != nil {
args.Fields.Resources = *resources
}
if taskStates != nil {
args.Fields.TaskStates = *taskStates
}
}
2015-09-06 22:37:21 +00:00
var out structs.AllocListResponse
if err := s.agent.RPC("Alloc.List", &args, &out); err != nil {
return nil, err
}
setMeta(resp, &out.QueryMeta)
2015-09-07 17:03:10 +00:00
if out.Allocations == nil {
out.Allocations = make([]*structs.AllocListStub, 0)
}
for _, alloc := range out.Allocations {
alloc.SetEventDisplayMessages()
}
2015-09-06 22:37:21 +00:00
return out.Allocations, nil
}
func (s *HTTPServer) AllocSpecificRequest(resp http.ResponseWriter, req *http.Request) (interface{}, error) {
reqSuffix := strings.TrimPrefix(req.URL.Path, "/v1/allocation/")
// tokenize the suffix of the path to get the alloc id and find the action
// invoked on the alloc id
tokens := strings.Split(reqSuffix, "/")
if len(tokens) > 2 || len(tokens) < 1 {
return nil, CodedError(404, resourceNotFoundErr)
}
allocID := tokens[0]
if len(tokens) == 1 {
return s.allocGet(allocID, resp, req)
}
switch tokens[1] {
case "checks":
return s.allocChecks(allocID, resp, req)
case "stop":
return s.allocStop(allocID, resp, req)
case "services":
return s.allocServiceRegistrations(resp, req, allocID)
}
return nil, CodedError(404, resourceNotFoundErr)
}
func (s *HTTPServer) allocGet(allocID string, resp http.ResponseWriter, req *http.Request) (interface{}, error) {
2015-09-06 22:49:44 +00:00
if req.Method != "GET" {
2015-09-06 22:37:21 +00:00
return nil, CodedError(405, ErrInvalidMethod)
}
2015-09-06 22:49:44 +00:00
args := structs.AllocSpecificRequest{
AllocID: allocID,
}
if s.parse(resp, req, &args.Region, &args.QueryOptions) {
return nil, nil
}
var out structs.SingleAllocResponse
if err := s.agent.RPC("Alloc.GetAlloc", &args, &out); err != nil {
return nil, err
}
2015-09-06 22:37:21 +00:00
2015-09-06 22:49:44 +00:00
setMeta(resp, &out.QueryMeta)
if out.Alloc == nil {
return nil, CodedError(404, "alloc not found")
}
2016-11-29 00:05:56 +00:00
2016-12-14 20:50:08 +00:00
// Decode the payload if there is any
2016-11-29 00:05:56 +00:00
alloc := out.Alloc
2016-12-14 20:50:08 +00:00
if alloc.Job != nil && len(alloc.Job.Payload) != 0 {
decoded, err := snappy.Decode(nil, alloc.Job.Payload)
2016-11-29 00:05:56 +00:00
if err != nil {
return nil, err
}
alloc = alloc.Copy()
2016-12-14 20:50:08 +00:00
alloc.Job.Payload = decoded
2016-11-29 00:05:56 +00:00
}
alloc.SetEventDisplayMessages()
2016-11-29 00:05:56 +00:00
// Handle 0.12 ports upgrade path
alloc = alloc.Copy()
alloc.AllocatedResources.Canonicalize()
2016-11-29 00:05:56 +00:00
return alloc, nil
2015-09-06 22:37:21 +00:00
}
2016-05-24 23:41:35 +00:00
func (s *HTTPServer) allocStop(allocID string, resp http.ResponseWriter, req *http.Request) (interface{}, error) {
if !(req.Method == "POST" || req.Method == "PUT") {
return nil, CodedError(405, ErrInvalidMethod)
}
2016-05-24 23:41:35 +00:00
noShutdownDelay := false
if noShutdownDelayQS := req.URL.Query().Get("no_shutdown_delay"); noShutdownDelayQS != "" {
var err error
noShutdownDelay, err = strconv.ParseBool(noShutdownDelayQS)
if err != nil {
return nil, fmt.Errorf("no_shutdown_delay value is not a boolean: %v", err)
}
}
sr := &structs.AllocStopRequest{
AllocID: allocID,
NoShutdownDelay: noShutdownDelay,
}
s.parseWriteRequest(req, &sr.WriteRequest)
var out structs.AllocStopResponse
rpcErr := s.agent.RPC("Alloc.Stop", &sr, &out)
if rpcErr != nil {
if structs.IsErrUnknownAllocation(rpcErr) {
rpcErr = CodedError(404, allocNotFoundErr)
}
return nil, rpcErr
}
setIndex(resp, out.Index)
return &out, nil
}
// allocServiceRegistrations returns a list of all service registrations
// assigned to the job identifier. It is callable via the
// /v1/allocation/:alloc_id/services HTTP API and uses the
// structs.AllocServiceRegistrationsRPCMethod RPC method.
func (s *HTTPServer) allocServiceRegistrations(
resp http.ResponseWriter, req *http.Request, allocID string) (interface{}, error) {
// The endpoint only supports GET requests.
if req.Method != http.MethodGet {
return nil, CodedError(http.StatusMethodNotAllowed, ErrInvalidMethod)
}
// Set up the request args and parse this to ensure the query options are
// set.
args := structs.AllocServiceRegistrationsRequest{AllocID: allocID}
if s.parse(resp, req, &args.Region, &args.QueryOptions) {
return nil, nil
}
// Perform the RPC request.
var reply structs.AllocServiceRegistrationsResponse
if err := s.agent.RPC(structs.AllocServiceRegistrationsRPCMethod, &args, &reply); err != nil {
return nil, err
}
setMeta(resp, &reply.QueryMeta)
if reply.Services == nil {
return nil, CodedError(http.StatusNotFound, allocNotFoundErr)
}
return reply.Services, nil
}
func (s *HTTPServer) ClientAllocRequest(resp http.ResponseWriter, req *http.Request) (interface{}, error) {
2016-05-24 23:41:35 +00:00
reqSuffix := strings.TrimPrefix(req.URL.Path, "/v1/client/allocation/")
// tokenize the suffix of the path to get the alloc id and find the action
// invoked on the alloc id
tokens := strings.Split(reqSuffix, "/")
if len(tokens) != 2 {
return nil, CodedError(404, resourceNotFoundErr)
2016-05-24 23:41:35 +00:00
}
allocID := tokens[0]
switch tokens[1] {
case "checks":
return s.allocChecks(allocID, resp, req)
case "stats":
return s.allocStats(allocID, resp, req)
case "exec":
return s.allocExec(allocID, resp, req)
case "snapshot":
2018-02-06 18:53:00 +00:00
if s.agent.client == nil {
return nil, clientNotRunning
}
2017-10-07 00:54:09 +00:00
return s.allocSnapshot(allocID, resp, req)
case "restart":
return s.allocRestart(allocID, resp, req)
case "gc":
return s.allocGC(allocID, resp, req)
case "signal":
return s.allocSignal(allocID, resp, req)
}
return nil, CodedError(404, resourceNotFoundErr)
}
func (s *HTTPServer) ClientGCRequest(resp http.ResponseWriter, req *http.Request) (interface{}, error) {
2018-02-06 18:53:00 +00:00
// Get the requested Node ID
requestedNode := req.URL.Query().Get("node_id")
// Build the request and parse the ACL token
args := structs.NodeSpecificRequest{
NodeID: requestedNode,
}
s.parse(resp, req, &args.QueryOptions.Region, &args.QueryOptions)
// Determine the handler to use
useLocalClient, useClientRPC, useServerRPC := s.rpcHandlerForNode(requestedNode)
// Make the RPC
var reply structs.GenericResponse
var rpcErr error
if useLocalClient {
rpcErr = s.agent.Client().ClientRPC("Allocations.GarbageCollectAll", &args, &reply)
} else if useClientRPC {
rpcErr = s.agent.Client().RPC("ClientAllocations.GarbageCollectAll", &args, &reply)
} else if useServerRPC {
rpcErr = s.agent.Server().RPC("ClientAllocations.GarbageCollectAll", &args, &reply)
} else {
rpcErr = CodedError(400, "No local Node and node_id not provided")
}
if rpcErr != nil {
if structs.IsErrNoNodeConn(rpcErr) {
rpcErr = CodedError(404, rpcErr.Error())
}
2017-10-04 22:08:58 +00:00
}
2018-02-06 18:53:00 +00:00
return nil, rpcErr
}
func (s *HTTPServer) allocRestart(allocID string, resp http.ResponseWriter, req *http.Request) (interface{}, error) {
// Build the request and parse the ACL token
args := structs.AllocRestartRequest{
AllocID: allocID,
TaskName: "",
}
s.parse(resp, req, &args.QueryOptions.Region, &args.QueryOptions)
// Explicitly parse the body separately to disallow overriding AllocID in req Body.
var reqBody struct {
TaskName string
Task lifecycle restart (#14127) * allocrunner: handle lifecycle when all tasks die When all tasks die the Coordinator must transition to its terminal state, coordinatorStatePoststop, to unblock poststop tasks. Since this could happen at any time (for example, a prestart task dies), all states must be able to transition to this terminal state. * allocrunner: implement different alloc restarts Add a new alloc restart mode where all tasks are restarted, even if they have already exited. Also unifies the alloc restart logic to use the implementation that restarts tasks concurrently and ignores ErrTaskNotRunning errors since those are expected when restarting the allocation. * allocrunner: allow tasks to run again Prevent the task runner Run() method from exiting to allow a dead task to run again. When the task runner is signaled to restart, the function will jump back to the MAIN loop and run it again. The task runner determines if a task needs to run again based on two new task events that were added to differentiate between a request to restart a specific task, the tasks that are currently running, or all tasks that have already run. * api/cli: add support for all tasks alloc restart Implement the new -all-tasks alloc restart CLI flag and its API counterpar, AllTasks. The client endpoint calls the appropriate restart method from the allocrunner depending on the restart parameters used. * test: fix tasklifecycle Coordinator test * allocrunner: kill taskrunners if all tasks are dead When all non-poststop tasks are dead we need to kill the taskrunners so we don't leak their goroutines, which are blocked in the alloc restart loop. This also ensures the allocrunner exits on its own. * taskrunner: fix tests that waited on WaitCh Now that "dead" tasks may run again, the taskrunner Run() method will not return when the task finishes running, so tests must wait for the task state to be "dead" instead of using the WaitCh, since it won't be closed until the taskrunner is killed. * tests: add tests for all tasks alloc restart * changelog: add entry for #14127 * taskrunner: fix restore logic. The first implementation of the task runner restore process relied on server data (`tr.Alloc().TerminalStatus()`) which may not be available to the client at the time of restore. It also had the incorrect code path. When restoring a dead task the driver handle always needs to be clear cleanly using `clearDriverHandle` otherwise, after exiting the MAIN loop, the task may be killed by `tr.handleKill`. The fix is to store the state of the Run() loop in the task runner local client state: if the task runner ever exits this loop cleanly (not with a shutdown) it will never be able to run again. So if the Run() loops starts with this local state flag set, it must exit early. This local state flag is also being checked on task restart requests. If the task is "dead" and its Run() loop is not active it will never be able to run again. * address code review requests * apply more code review changes * taskrunner: add different Restart modes Using the task event to differentiate between the allocrunner restart methods proved to be confusing for developers to understand how it all worked. So instead of relying on the event type, this commit separated the logic of restarting an taskRunner into two methods: - `Restart` will retain the current behaviour and only will only restart the task if it's currently running. - `ForceRestart` is the new method where a `dead` task is allowed to restart if its `Run()` method is still active. Callers will need to restart the allocRunner taskCoordinator to make sure it will allow the task to run again. * minor fixes
2022-08-24 21:43:07 +00:00
AllTasks bool
}
err := json.NewDecoder(req.Body).Decode(&reqBody)
if err != nil && err != io.EOF {
return nil, err
}
if reqBody.TaskName != "" {
args.TaskName = reqBody.TaskName
}
Task lifecycle restart (#14127) * allocrunner: handle lifecycle when all tasks die When all tasks die the Coordinator must transition to its terminal state, coordinatorStatePoststop, to unblock poststop tasks. Since this could happen at any time (for example, a prestart task dies), all states must be able to transition to this terminal state. * allocrunner: implement different alloc restarts Add a new alloc restart mode where all tasks are restarted, even if they have already exited. Also unifies the alloc restart logic to use the implementation that restarts tasks concurrently and ignores ErrTaskNotRunning errors since those are expected when restarting the allocation. * allocrunner: allow tasks to run again Prevent the task runner Run() method from exiting to allow a dead task to run again. When the task runner is signaled to restart, the function will jump back to the MAIN loop and run it again. The task runner determines if a task needs to run again based on two new task events that were added to differentiate between a request to restart a specific task, the tasks that are currently running, or all tasks that have already run. * api/cli: add support for all tasks alloc restart Implement the new -all-tasks alloc restart CLI flag and its API counterpar, AllTasks. The client endpoint calls the appropriate restart method from the allocrunner depending on the restart parameters used. * test: fix tasklifecycle Coordinator test * allocrunner: kill taskrunners if all tasks are dead When all non-poststop tasks are dead we need to kill the taskrunners so we don't leak their goroutines, which are blocked in the alloc restart loop. This also ensures the allocrunner exits on its own. * taskrunner: fix tests that waited on WaitCh Now that "dead" tasks may run again, the taskrunner Run() method will not return when the task finishes running, so tests must wait for the task state to be "dead" instead of using the WaitCh, since it won't be closed until the taskrunner is killed. * tests: add tests for all tasks alloc restart * changelog: add entry for #14127 * taskrunner: fix restore logic. The first implementation of the task runner restore process relied on server data (`tr.Alloc().TerminalStatus()`) which may not be available to the client at the time of restore. It also had the incorrect code path. When restoring a dead task the driver handle always needs to be clear cleanly using `clearDriverHandle` otherwise, after exiting the MAIN loop, the task may be killed by `tr.handleKill`. The fix is to store the state of the Run() loop in the task runner local client state: if the task runner ever exits this loop cleanly (not with a shutdown) it will never be able to run again. So if the Run() loops starts with this local state flag set, it must exit early. This local state flag is also being checked on task restart requests. If the task is "dead" and its Run() loop is not active it will never be able to run again. * address code review requests * apply more code review changes * taskrunner: add different Restart modes Using the task event to differentiate between the allocrunner restart methods proved to be confusing for developers to understand how it all worked. So instead of relying on the event type, this commit separated the logic of restarting an taskRunner into two methods: - `Restart` will retain the current behaviour and only will only restart the task if it's currently running. - `ForceRestart` is the new method where a `dead` task is allowed to restart if its `Run()` method is still active. Callers will need to restart the allocRunner taskCoordinator to make sure it will allow the task to run again. * minor fixes
2022-08-24 21:43:07 +00:00
if reqBody.AllTasks {
args.AllTasks = reqBody.AllTasks
}
// Determine the handler to use
useLocalClient, useClientRPC, useServerRPC := s.rpcHandlerForAlloc(allocID)
// Make the RPC
var reply structs.GenericResponse
var rpcErr error
if useLocalClient {
rpcErr = s.agent.Client().ClientRPC("Allocations.Restart", &args, &reply)
} else if useClientRPC {
rpcErr = s.agent.Client().RPC("ClientAllocations.Restart", &args, &reply)
} else if useServerRPC {
rpcErr = s.agent.Server().RPC("ClientAllocations.Restart", &args, &reply)
} else {
rpcErr = CodedError(400, "No local Node and node_id not provided")
}
if rpcErr != nil {
if structs.IsErrNoNodeConn(rpcErr) || structs.IsErrUnknownAllocation(rpcErr) {
rpcErr = CodedError(404, rpcErr.Error())
}
}
return reply, rpcErr
}
func (s *HTTPServer) allocGC(allocID string, resp http.ResponseWriter, req *http.Request) (interface{}, error) {
2018-02-06 18:53:00 +00:00
// Build the request and parse the ACL token
args := structs.AllocSpecificRequest{
AllocID: allocID,
}
s.parse(resp, req, &args.QueryOptions.Region, &args.QueryOptions)
2018-02-06 18:53:00 +00:00
// Determine the handler to use
useLocalClient, useClientRPC, useServerRPC := s.rpcHandlerForAlloc(allocID)
2018-02-06 18:53:00 +00:00
// Make the RPC
var reply structs.GenericResponse
var rpcErr error
if useLocalClient {
rpcErr = s.agent.Client().ClientRPC("Allocations.GarbageCollect", &args, &reply)
} else if useClientRPC {
rpcErr = s.agent.Client().RPC("ClientAllocations.GarbageCollect", &args, &reply)
} else if useServerRPC {
rpcErr = s.agent.Server().RPC("ClientAllocations.GarbageCollect", &args, &reply)
} else {
rpcErr = CodedError(400, "No local Node and node_id not provided")
}
2018-02-06 18:53:00 +00:00
if rpcErr != nil {
2018-02-13 23:50:51 +00:00
if structs.IsErrNoNodeConn(rpcErr) || structs.IsErrUnknownAllocation(rpcErr) {
2018-02-06 18:53:00 +00:00
rpcErr = CodedError(404, rpcErr.Error())
}
}
2018-02-06 18:53:00 +00:00
return nil, rpcErr
}
func (s *HTTPServer) allocSignal(allocID string, resp http.ResponseWriter, req *http.Request) (interface{}, error) {
if !(req.Method == "POST" || req.Method == "PUT") {
return nil, CodedError(405, ErrInvalidMethod)
}
// Build the request and parse the ACL token
args := structs.AllocSignalRequest{}
err := decodeBody(req, &args)
if err != nil {
return nil, CodedError(400, fmt.Sprintf("Failed to decode body: %v", err))
}
s.parse(resp, req, &args.QueryOptions.Region, &args.QueryOptions)
args.AllocID = allocID
// Determine the handler to use
useLocalClient, useClientRPC, useServerRPC := s.rpcHandlerForAlloc(allocID)
// Make the RPC
var reply structs.GenericResponse
var rpcErr error
if useLocalClient {
rpcErr = s.agent.Client().ClientRPC("Allocations.Signal", &args, &reply)
} else if useClientRPC {
rpcErr = s.agent.Client().RPC("ClientAllocations.Signal", &args, &reply)
} else if useServerRPC {
rpcErr = s.agent.Server().RPC("ClientAllocations.Signal", &args, &reply)
} else {
rpcErr = CodedError(400, "No local Node and node_id not provided")
}
if rpcErr != nil {
if structs.IsErrNoNodeConn(rpcErr) || structs.IsErrUnknownAllocation(rpcErr) {
rpcErr = CodedError(404, rpcErr.Error())
}
}
return reply, rpcErr
}
2017-10-07 00:54:09 +00:00
func (s *HTTPServer) allocSnapshot(allocID string, resp http.ResponseWriter, req *http.Request) (interface{}, error) {
var secret string
s.parseToken(req, &secret)
if !s.agent.Client().ValidateMigrateToken(allocID, secret) {
return nil, structs.ErrPermissionDenied
}
allocFS, err := s.agent.Client().GetAllocFS(allocID)
if err != nil {
return nil, fmt.Errorf(allocNotFoundErr)
}
if err := allocFS.Snapshot(resp); err != nil {
return nil, fmt.Errorf("error making snapshot: %v", err)
}
return nil, nil
}
2016-05-24 23:41:35 +00:00
func (s *HTTPServer) allocStats(allocID string, resp http.ResponseWriter, req *http.Request) (interface{}, error) {
2018-02-06 18:53:00 +00:00
// Build the request and parse the ACL token
task := req.URL.Query().Get("task")
args := cstructs.AllocStatsRequest{
AllocID: allocID,
Task: task,
}
s.parse(resp, req, &args.QueryOptions.Region, &args.QueryOptions)
2018-02-06 18:53:00 +00:00
// Determine the handler to use
useLocalClient, useClientRPC, useServerRPC := s.rpcHandlerForAlloc(allocID)
// Make the RPC
var reply cstructs.AllocStatsResponse
var rpcErr error
if useLocalClient {
rpcErr = s.agent.Client().ClientRPC("Allocations.Stats", &args, &reply)
} else if useClientRPC {
rpcErr = s.agent.Client().RPC("ClientAllocations.Stats", &args, &reply)
} else if useServerRPC {
rpcErr = s.agent.Server().RPC("ClientAllocations.Stats", &args, &reply)
} else {
rpcErr = CodedError(400, "No local Node and node_id not provided")
}
2018-02-06 18:53:00 +00:00
if rpcErr != nil {
2018-02-13 23:50:51 +00:00
if structs.IsErrNoNodeConn(rpcErr) || structs.IsErrUnknownAllocation(rpcErr) {
2018-02-06 18:53:00 +00:00
rpcErr = CodedError(404, rpcErr.Error())
}
}
2018-02-06 18:53:00 +00:00
return reply.Stats, rpcErr
2016-05-24 23:41:35 +00:00
}
func (s *HTTPServer) allocChecks(allocID string, resp http.ResponseWriter, req *http.Request) (any, error) {
// Build the request and parse the ACL token
args := cstructs.AllocChecksRequest{
AllocID: allocID,
}
s.parse(resp, req, &args.QueryOptions.Region, &args.QueryOptions)
// Determine the handler to use
useLocalClient, useClientRPC, useServerRPC := s.rpcHandlerForAlloc(allocID)
// Make the RPC
var reply cstructs.AllocChecksResponse
var rpcErr error
switch {
case useLocalClient:
rpcErr = s.agent.Client().ClientRPC("Allocations.Checks", &args, &reply)
case useClientRPC:
rpcErr = s.agent.Client().RPC("Allocations.Checks", &args, &reply)
case useServerRPC:
rpcErr = s.agent.Server().RPC("Allocations.Checks", &args, &reply)
default:
rpcErr = CodedError(400, "No local Node and node_id not provided")
}
if rpcErr != nil {
if structs.IsErrNoNodeConn(rpcErr) || structs.IsErrUnknownAllocation(rpcErr) {
rpcErr = CodedError(404, rpcErr.Error())
}
}
return reply.Results, rpcErr
}
func (s *HTTPServer) allocExec(allocID string, resp http.ResponseWriter, req *http.Request) (interface{}, error) {
// Build the request and parse the ACL token
task := req.URL.Query().Get("task")
cmdJsonStr := req.URL.Query().Get("command")
var command []string
err := json.Unmarshal([]byte(cmdJsonStr), &command)
if err != nil {
// this shouldn't happen, []string is always be serializable to json
return nil, fmt.Errorf("failed to marshal command into json: %v", err)
}
ttyB := false
if tty := req.URL.Query().Get("tty"); tty != "" {
ttyB, err = strconv.ParseBool(tty)
if err != nil {
return nil, fmt.Errorf("tty value is not a boolean: %v", err)
}
}
args := cstructs.AllocExecRequest{
AllocID: allocID,
Task: task,
Cmd: command,
Tty: ttyB,
}
s.parse(resp, req, &args.QueryOptions.Region, &args.QueryOptions)
conn, err := s.wsUpgrader.Upgrade(resp, req, nil)
if err != nil {
return nil, fmt.Errorf("failed to upgrade connection: %v", err)
}
if err := readWsHandshake(conn.ReadJSON, req, &args.QueryOptions); err != nil {
conn.WriteMessage(websocket.CloseMessage,
websocket.FormatCloseMessage(toWsCode(400), err.Error()))
return nil, err
}
return s.execStreamImpl(conn, &args)
}
// readWsHandshake reads the websocket handshake message and sets
// query authentication token, if request requires a handshake
func readWsHandshake(readFn func(interface{}) error, req *http.Request, q *structs.QueryOptions) error {
// Avoid handshake if request doesn't require one
if hv := req.URL.Query().Get("ws_handshake"); hv == "" {
return nil
} else if h, err := strconv.ParseBool(hv); err != nil {
return fmt.Errorf("ws_handshake value is not a boolean: %v", err)
} else if !h {
return nil
}
var h wsHandshakeMessage
err := readFn(&h)
if err != nil {
return err
}
supportedWSHandshakeVersion := 1
if h.Version != supportedWSHandshakeVersion {
return fmt.Errorf("unexpected handshake value: %v", h.Version)
}
q.AuthToken = h.AuthToken
return nil
}
type wsHandshakeMessage struct {
Version int `json:"version"`
AuthToken string `json:"auth_token"`
}
func (s *HTTPServer) execStreamImpl(ws *websocket.Conn, args *cstructs.AllocExecRequest) (interface{}, error) {
allocID := args.AllocID
method := "Allocations.Exec"
// Get the correct handler
localClient, remoteClient, localServer := s.rpcHandlerForAlloc(allocID)
var handler structs.StreamingRpcHandler
var handlerErr error
if localClient {
handler, handlerErr = s.agent.Client().StreamingRpcHandler(method)
} else if remoteClient {
handler, handlerErr = s.agent.Client().RemoteStreamingRpcHandler(method)
} else if localServer {
handler, handlerErr = s.agent.Server().StreamingRpcHandler(method)
}
if handlerErr != nil {
return nil, CodedError(500, handlerErr.Error())
}
// Create a pipe connecting the (possibly remote) handler to the http response
httpPipe, handlerPipe := net.Pipe()
decoder := codec.NewDecoder(httpPipe, structs.MsgpackHandle)
encoder := codec.NewEncoder(httpPipe, structs.MsgpackHandle)
// Create a goroutine that closes the pipe if the connection closes.
ctx, cancel := context.WithCancel(context.Background())
go func() {
<-ctx.Done()
httpPipe.Close()
// don't close ws - wait to drain messages
}()
// Create a channel that decodes the results
errCh := make(chan HTTPCodedError, 2)
// stream response
go func() {
defer cancel()
// Send the request
if err := encoder.Encode(args); err != nil {
errCh <- CodedError(500, err.Error())
return
}
go forwardExecInput(encoder, ws, errCh)
for {
var res cstructs.StreamErrWrapper
err := decoder.Decode(&res)
if isClosedError(err) {
ws.WriteMessage(websocket.CloseMessage, websocket.FormatCloseMessage(websocket.CloseNormalClosure, ""))
errCh <- nil
return
}
if err != nil {
errCh <- CodedError(500, err.Error())
return
}
decoder.Reset(httpPipe)
if err := res.Error; err != nil {
code := 500
if err.Code != nil {
code = int(*err.Code)
}
errCh <- CodedError(code, err.Error())
return
}
if err := ws.WriteMessage(websocket.TextMessage, res.Payload); err != nil {
errCh <- CodedError(500, err.Error())
return
}
}
}()
// start streaming request to streaming RPC - returns when streaming completes or errors
handler(handlerPipe)
// stop streaming background goroutines for streaming - but not websocket activity
cancel()
// retrieve any error and/or wait until goroutine stop and close errCh connection before
// closing websocket connection
codedErr := <-errCh
// we won't return an error on ws close, but at least make it available in
// the logs so we can trace spurious disconnects
if codedErr != nil {
s.logger.Debug("alloc exec channel closed with error", "error", codedErr)
}
if isClosedError(codedErr) {
codedErr = nil
} else if codedErr != nil {
ws.WriteMessage(websocket.CloseMessage,
websocket.FormatCloseMessage(toWsCode(codedErr.Code()), codedErr.Error()))
}
ws.Close()
return nil, codedErr
}
func toWsCode(httpCode int) int {
switch httpCode {
case 500:
return websocket.CloseInternalServerErr
default:
// placeholder error code
return websocket.ClosePolicyViolation
}
}
func isClosedError(err error) bool {
if err == nil {
return false
}
return err == io.EOF ||
err == io.ErrClosedPipe ||
strings.Contains(err.Error(), "closed") ||
strings.Contains(err.Error(), "EOF")
}
// forwardExecInput forwards exec input (e.g. stdin) from websocket connection
// to the streaming RPC connection to client
func forwardExecInput(encoder *codec.Encoder, ws *websocket.Conn, errCh chan<- HTTPCodedError) {
for {
sf := &drivers.ExecTaskStreamingRequestMsg{}
err := ws.ReadJSON(sf)
if err == io.EOF {
return
}
if err != nil {
errCh <- CodedError(500, err.Error())
return
}
err = encoder.Encode(sf)
if err != nil {
errCh <- CodedError(500, err.Error())
}
}
}