open-nomad/command/agent/eval_endpoint.go

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

170 lines
4.6 KiB
Go
Raw Permalink Normal View History

// Copyright (c) HashiCorp, Inc.
// SPDX-License-Identifier: MPL-2.0
2015-09-06 23:02:53 +00:00
package agent
import (
"fmt"
2015-09-06 23:02:53 +00:00
"net/http"
"strings"
"github.com/hashicorp/nomad/nomad/structs"
)
// EvalsRequest is the entry point for /v1/evaluations and is responsible for
// handling both the listing of evaluations, and the bulk deletion of
// evaluations. The latter is a dangerous operation and should use the
// eval delete command to perform this.
2015-09-06 23:02:53 +00:00
func (s *HTTPServer) EvalsRequest(resp http.ResponseWriter, req *http.Request) (interface{}, error) {
switch req.Method {
case http.MethodGet:
return s.evalsListRequest(resp, req)
case http.MethodDelete:
return s.evalsDeleteRequest(resp, req)
default:
return nil, CodedError(http.StatusMethodNotAllowed, ErrInvalidMethod)
2015-09-06 23:02:53 +00:00
}
}
func (s *HTTPServer) evalsListRequest(resp http.ResponseWriter, req *http.Request) (interface{}, error) {
2015-09-06 23:02:53 +00:00
args := structs.EvalListRequest{}
if s.parse(resp, req, &args.Region, &args.QueryOptions) {
return nil, nil
}
query := req.URL.Query()
args.FilterEvalStatus = query.Get("status")
args.FilterJobID = query.Get("job")
2015-09-06 23:02:53 +00:00
var out structs.EvalListResponse
if err := s.agent.RPC("Eval.List", &args, &out); err != nil {
return nil, err
}
setMeta(resp, &out.QueryMeta)
2015-09-07 17:03:10 +00:00
if out.Evaluations == nil {
out.Evaluations = make([]*structs.Evaluation, 0)
}
2015-09-06 23:02:53 +00:00
return out.Evaluations, nil
}
func (s *HTTPServer) evalsDeleteRequest(resp http.ResponseWriter, req *http.Request) (interface{}, error) {
var args structs.EvalDeleteRequest
if err := decodeBody(req, &args); err != nil {
return nil, CodedError(http.StatusBadRequest, err.Error())
}
numIDs := len(args.EvalIDs)
eval delete: move batching of deletes into RPC handler and state (#15117) During unusual outage recovery scenarios on large clusters, a backlog of millions of evaluations can appear. In these cases, the `eval delete` command can put excessive load on the cluster by listing large sets of evals to extract the IDs and then sending larges batches of IDs. Although the command's batch size was carefully tuned, we still need to be JSON deserialize, re-serialize to MessagePack, send the log entries through raft, and get the FSM applied. To improve performance of this recovery case, move the batching process into the RPC handler and the state store. The design here is a little weird, so let's look a the failed options first: * A naive solution here would be to just send the filter as the raft request and let the FSM apply delete the whole set in a single operation. Benchmarking with 1M evals on a 3 node cluster demonstrated this can block the FSM apply for several minutes, which puts the cluster at risk if there's a leadership failover (the barrier write can't be made while this apply is in-flight). * A less naive but still bad solution would be to have the RPC handler filter and paginate, and then hand a list of IDs to the existing raft log entry. Benchmarks showed this blocked the FSM apply for 20-30s at a time and took roughly an hour to complete. Instead, we're filtering and paginating in the RPC handler to find a page token, and then passing both the filter and page token in the raft log. The FSM apply recreates the paginator using the filter and page token to get roughly the same page of evaluations, which it then deletes. The pagination process is fairly cheap (only abut 5% of the total FSM apply time), so counter-intuitively this rework ends up being much faster. A benchmark of 1M evaluations showed this blocked the FSM apply for 20-30ms at a time (typical for normal operations) and completes in less than 4 minutes. Note that, as with the existing design, this delete is not consistent: a new evaluation inserted "behind" the cursor of the pagination will fail to be deleted.
2022-11-14 19:08:13 +00:00
if args.Filter != "" && numIDs > 0 {
return nil, CodedError(http.StatusBadRequest,
"evals cannot be deleted by both ID and filter")
}
if args.Filter == "" && numIDs == 0 {
return nil, CodedError(http.StatusBadRequest,
"evals must be deleted by either ID or filter")
}
// If an explicit list of evaluation IDs is sent, ensure its within bounds
if numIDs > structs.MaxUUIDsPerWriteRequest {
return nil, CodedError(http.StatusBadRequest, fmt.Sprintf(
eval delete: move batching of deletes into RPC handler and state (#15117) During unusual outage recovery scenarios on large clusters, a backlog of millions of evaluations can appear. In these cases, the `eval delete` command can put excessive load on the cluster by listing large sets of evals to extract the IDs and then sending larges batches of IDs. Although the command's batch size was carefully tuned, we still need to be JSON deserialize, re-serialize to MessagePack, send the log entries through raft, and get the FSM applied. To improve performance of this recovery case, move the batching process into the RPC handler and the state store. The design here is a little weird, so let's look a the failed options first: * A naive solution here would be to just send the filter as the raft request and let the FSM apply delete the whole set in a single operation. Benchmarking with 1M evals on a 3 node cluster demonstrated this can block the FSM apply for several minutes, which puts the cluster at risk if there's a leadership failover (the barrier write can't be made while this apply is in-flight). * A less naive but still bad solution would be to have the RPC handler filter and paginate, and then hand a list of IDs to the existing raft log entry. Benchmarks showed this blocked the FSM apply for 20-30s at a time and took roughly an hour to complete. Instead, we're filtering and paginating in the RPC handler to find a page token, and then passing both the filter and page token in the raft log. The FSM apply recreates the paginator using the filter and page token to get roughly the same page of evaluations, which it then deletes. The pagination process is fairly cheap (only abut 5% of the total FSM apply time), so counter-intuitively this rework ends up being much faster. A benchmark of 1M evaluations showed this blocked the FSM apply for 20-30ms at a time (typical for normal operations) and completes in less than 4 minutes. Note that, as with the existing design, this delete is not consistent: a new evaluation inserted "behind" the cursor of the pagination will fail to be deleted.
2022-11-14 19:08:13 +00:00
"request includes %v evaluation IDs, must be %v or fewer",
numIDs, structs.MaxUUIDsPerWriteRequest))
}
// Pass the write request to populate all meta fields.
s.parseWriteRequest(req, &args.WriteRequest)
var reply structs.EvalDeleteResponse
if err := s.agent.RPC(structs.EvalDeleteRPCMethod, &args, &reply); err != nil {
return nil, err
}
eval delete: move batching of deletes into RPC handler and state (#15117) During unusual outage recovery scenarios on large clusters, a backlog of millions of evaluations can appear. In these cases, the `eval delete` command can put excessive load on the cluster by listing large sets of evals to extract the IDs and then sending larges batches of IDs. Although the command's batch size was carefully tuned, we still need to be JSON deserialize, re-serialize to MessagePack, send the log entries through raft, and get the FSM applied. To improve performance of this recovery case, move the batching process into the RPC handler and the state store. The design here is a little weird, so let's look a the failed options first: * A naive solution here would be to just send the filter as the raft request and let the FSM apply delete the whole set in a single operation. Benchmarking with 1M evals on a 3 node cluster demonstrated this can block the FSM apply for several minutes, which puts the cluster at risk if there's a leadership failover (the barrier write can't be made while this apply is in-flight). * A less naive but still bad solution would be to have the RPC handler filter and paginate, and then hand a list of IDs to the existing raft log entry. Benchmarks showed this blocked the FSM apply for 20-30s at a time and took roughly an hour to complete. Instead, we're filtering and paginating in the RPC handler to find a page token, and then passing both the filter and page token in the raft log. The FSM apply recreates the paginator using the filter and page token to get roughly the same page of evaluations, which it then deletes. The pagination process is fairly cheap (only abut 5% of the total FSM apply time), so counter-intuitively this rework ends up being much faster. A benchmark of 1M evaluations showed this blocked the FSM apply for 20-30ms at a time (typical for normal operations) and completes in less than 4 minutes. Note that, as with the existing design, this delete is not consistent: a new evaluation inserted "behind" the cursor of the pagination will fail to be deleted.
2022-11-14 19:08:13 +00:00
setIndex(resp, reply.Index)
eval delete: move batching of deletes into RPC handler and state (#15117) During unusual outage recovery scenarios on large clusters, a backlog of millions of evaluations can appear. In these cases, the `eval delete` command can put excessive load on the cluster by listing large sets of evals to extract the IDs and then sending larges batches of IDs. Although the command's batch size was carefully tuned, we still need to be JSON deserialize, re-serialize to MessagePack, send the log entries through raft, and get the FSM applied. To improve performance of this recovery case, move the batching process into the RPC handler and the state store. The design here is a little weird, so let's look a the failed options first: * A naive solution here would be to just send the filter as the raft request and let the FSM apply delete the whole set in a single operation. Benchmarking with 1M evals on a 3 node cluster demonstrated this can block the FSM apply for several minutes, which puts the cluster at risk if there's a leadership failover (the barrier write can't be made while this apply is in-flight). * A less naive but still bad solution would be to have the RPC handler filter and paginate, and then hand a list of IDs to the existing raft log entry. Benchmarks showed this blocked the FSM apply for 20-30s at a time and took roughly an hour to complete. Instead, we're filtering and paginating in the RPC handler to find a page token, and then passing both the filter and page token in the raft log. The FSM apply recreates the paginator using the filter and page token to get roughly the same page of evaluations, which it then deletes. The pagination process is fairly cheap (only abut 5% of the total FSM apply time), so counter-intuitively this rework ends up being much faster. A benchmark of 1M evaluations showed this blocked the FSM apply for 20-30ms at a time (typical for normal operations) and completes in less than 4 minutes. Note that, as with the existing design, this delete is not consistent: a new evaluation inserted "behind" the cursor of the pagination will fail to be deleted.
2022-11-14 19:08:13 +00:00
return reply, nil
}
2015-09-06 23:02:53 +00:00
func (s *HTTPServer) EvalSpecificRequest(resp http.ResponseWriter, req *http.Request) (interface{}, error) {
2015-09-06 23:18:25 +00:00
path := strings.TrimPrefix(req.URL.Path, "/v1/evaluation/")
switch {
case strings.HasSuffix(path, "/allocations"):
evalID := strings.TrimSuffix(path, "/allocations")
return s.evalAllocations(resp, req, evalID)
default:
return s.evalQuery(resp, req, path)
}
}
func (s *HTTPServer) evalAllocations(resp http.ResponseWriter, req *http.Request, evalID string) (interface{}, error) {
if req.Method != http.MethodGet {
2015-09-06 23:18:25 +00:00
return nil, CodedError(405, ErrInvalidMethod)
}
args := structs.EvalSpecificRequest{
EvalID: evalID,
}
if s.parse(resp, req, &args.Region, &args.QueryOptions) {
return nil, nil
}
var out structs.EvalAllocationsResponse
if err := s.agent.RPC("Eval.Allocations", &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)
}
2015-09-06 23:18:25 +00:00
return out.Allocations, nil
}
func (s *HTTPServer) evalQuery(resp http.ResponseWriter, req *http.Request, evalID string) (interface{}, error) {
if req.Method != http.MethodGet {
2015-09-06 23:02:53 +00:00
return nil, CodedError(405, ErrInvalidMethod)
}
args := structs.EvalSpecificRequest{
EvalID: evalID,
}
if s.parse(resp, req, &args.Region, &args.QueryOptions) {
return nil, nil
}
query := req.URL.Query()
args.IncludeRelated = query.Get("related") == "true"
2015-09-06 23:02:53 +00:00
var out structs.SingleEvalResponse
if err := s.agent.RPC("Eval.GetEval", &args, &out); err != nil {
return nil, err
}
setMeta(resp, &out.QueryMeta)
if out.Eval == nil {
return nil, CodedError(404, "eval not found")
}
return out.Eval, nil
}
func (s *HTTPServer) EvalsCountRequest(resp http.ResponseWriter, req *http.Request) (interface{}, error) {
if req.Method != http.MethodGet {
return nil, CodedError(http.StatusMethodNotAllowed, ErrInvalidMethod)
}
args := structs.EvalCountRequest{}
if s.parse(resp, req, &args.Region, &args.QueryOptions) {
return nil, nil
}
var out structs.EvalCountResponse
if err := s.agent.RPC("Eval.Count", &args, &out); err != nil {
return nil, err
}
setMeta(resp, &out.QueryMeta)
return &out, nil
}