24666372903

Committed 20 Apr 2026 12:21PM UTC coverage: 82.697% (+0.07%) from 82.629%

Build # 24666372903

Build Type

Pull #260

github

Committed by

robertodauria

Commit Message

refactor(handler): adapt to m-lab/access variadic Sign API

Pull Request Pull Request #260: feat(handler): propagate int_id/key_id into priority access tokens

Coverage Stats

43 of 44 new or added lines in 3 files covered. (97.73%)

24 existing lines in 1 file now uncovered.

2514 of 3040 relevant lines covered (82.7%)

6.17 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

88.94

/handler/handler.go

// Package handler provides a client and handlers for responding to locate
// requests.
package handler

import (
        "bytes"
        "context"
        "encoding/json"
        "errors"
        "fmt"
        "html/template"
        "math/rand"
        "net"
        "net/http"
        "net/url"
        "path"
        "strconv"
        "strings"
        "time"

        "github.com/go-jose/go-jose/v4/jwt"
        "github.com/google/uuid"
        log "github.com/sirupsen/logrus"

        "github.com/m-lab/go/rtx"
        v2 "github.com/m-lab/locate/api/v2"
        "github.com/m-lab/locate/clientgeo"
        "github.com/m-lab/locate/heartbeat"
        "github.com/m-lab/locate/limits"
        "github.com/m-lab/locate/metrics"
        "github.com/m-lab/locate/siteinfo"
        "github.com/m-lab/locate/static"
        prom "github.com/prometheus/client_golang/api/prometheus/v1"
        "github.com/prometheus/common/model"
)

var (
        errFailedToLookupClient = errors.New("Failed to look up client location")
        tooManyRequests         = "Too many periodic requests. Please contact support@measurementlab.net."
)

// Signer defines how access tokens are signed. Extra claim objects are merged
// into the JWT payload via go-jose's variadic Claims support (see
// token.Signer.Sign).
type Signer interface {
        Sign(cl jwt.Claims, extra ...any) (string, error)
}

// IntegrationClaims holds M-Lab integration-specific claims that are embedded
// into access tokens issued for priority requests. The JSON tags are the
// wire-level claim names in the resulting JWT payload.
type IntegrationClaims struct {
        IntegrationID string `json:"int_id,omitempty"`
        KeyID         string `json:"key_id,omitempty"`
}

type Limiter interface {
        IsLimited(ip, ua string) (limits.LimitStatus, error)
}

// TierLimiter extends Limiter with tier-based limits for priority endpoints.
type TierLimiter interface {
        Limiter
        IsLimitedWithTier(org, ip string, tierConfig limits.LimitConfig) (limits.LimitStatus, error)
}

// Client handles HTTP requests for the locate service, including nearest server
// lookups, heartbeat processing, and monitoring endpoints. It maintains dependencies
// for geolocation, rate limiting, access token signing, and metrics collection.
// TODO: This should probably be a Handler, not a Client.
type Client struct {
        Signer
        project string
        LocatorV2
        ClientLocator
        PrometheusClient
        targetTmpl       *template.Template
        agentLimits      limits.Agents
        tierLimits       limits.TierLimits
        ipLimiter        TierLimiter
        earlyExitClients map[string]bool
        jwtVerifier      Verifier
}

// LocatorV2 defines how the Nearest handler requests machines nearest to the
// client.
type LocatorV2 interface {
        Nearest(service string, lat, lon float64, opts *heartbeat.NearestOptions) (*heartbeat.TargetInfo, error)
        heartbeat.StatusTracker
}

// ClientLocator defines the interface for looking up the client geolocation.
type ClientLocator interface {
        Locate(req *http.Request) (*clientgeo.Location, error)
}

// PrometheusClient defines the interface to query Prometheus.
type PrometheusClient interface {
        Query(ctx context.Context, query string, ts time.Time, opts ...prom.Option) (model.Value, prom.Warnings, error)
}

type paramOpts struct {
        raw       url.Values
        version   string
        ranks     map[string]int
        svcParams map[string]float64
}

func init() {
        log.SetFormatter(&log.JSONFormatter{})
        log.SetLevel(log.InfoLevel)
}

// NewClient creates a new client.
func NewClient(project string, private Signer, locatorV2 LocatorV2, client ClientLocator,
        promClient PrometheusClient, lmts limits.Agents, tierLmts limits.TierLimits, limiter TierLimiter, earlyExitClients []string, jwtVerifier Verifier) *Client {
        // Convert slice to map for O(1) lookups
        earlyExitMap := make(map[string]bool)
        for _, client := range earlyExitClients {
                earlyExitMap[client] = true
        }
        return &Client{
                Signer:           private,
                project:          project,
                LocatorV2:        locatorV2,
                ClientLocator:    client,
                PrometheusClient: promClient,
                targetTmpl:       template.Must(template.New("name").Parse("{{.Hostname}}{{.Ports}}")),
                agentLimits:      lmts,
                tierLimits:       tierLmts,
                ipLimiter:        limiter,
                earlyExitClients: earlyExitMap,
                jwtVerifier:      jwtVerifier,
        }
}

func (c *Client) extraParams(hostname string, index int, p paramOpts) url.Values {
        v := url.Values{}

        // Add client parameters.
        for key := range p.raw {
                if strings.HasPrefix(key, "client_") {
                        // note: we only use the first value.
                        v.Set(key, p.raw.Get(key))
                }

                val, ok := p.svcParams[key]
                if ok && rand.Float64() < val {
                        v.Set(key, p.raw.Get(key))
                }
        }

        // Add early_exit parameter for specified clients
        clientName := p.raw.Get("client_name")
        if clientName != "" && c.earlyExitClients[clientName] {
                v.Set(static.EarlyExitParameter, static.EarlyExitDefaultValue)
        }

        // Add Locate Service version.
        v.Set("locate_version", p.version)

        // Add metro rank.
        rank, ok := p.ranks[hostname]
        if ok {
                v.Set("metro_rank", strconv.Itoa(rank))
        }

        // Add result index.
        v.Set("index", strconv.Itoa(index))

        return v
}

// Nearest uses an implementation of the LocatorV2 interface to look up
// nearest servers.
func (c *Client) Nearest(rw http.ResponseWriter, req *http.Request) {
        result := v2.NearestResult{}
        setHeaders(rw)

        if err := req.ParseForm(); err != nil {
                log.Debugf("ParseForm error: %v", err)
                result.Error = v2.NewError("server", "Failed to parse form", http.StatusInternalServerError)
                writeResult(rw, result.Error.Status, result)
                metrics.RequestsTotal.WithLabelValues("nearest", "server",
                        http.StatusText(result.Error.Status)).Inc()
                return
        }

        if c.limitRequest(time.Now().UTC(), req) {
                result.Error = v2.NewError("client", tooManyRequests, http.StatusTooManyRequests)
                writeResult(rw, result.Error.Status, &result)
                metrics.RequestsTotal.WithLabelValues("nearest", "request limit", http.StatusText(result.Error.Status)).Inc()
                return
        }

        // Check rate limit for IP and UA.
        if c.ipLimiter != nil {
                ip := getRemoteAddr(req)
                if ip != "" {
                        // An empty UA is technically possible.
                        ua := req.Header.Get("User-Agent")
                        status, err := c.ipLimiter.IsLimited(ip, ua)
                        if err != nil {
                                // Log error but don't block request (fail open).
                                // TODO: Add tests for this path.
                                log.Printf("Rate limiter error: %v", err)
                        } else if status.IsLimited {
                                // Log IP and UA and block the request.
                                result.Error = v2.NewError("client", tooManyRequests, http.StatusTooManyRequests)
                                metrics.RequestsTotal.WithLabelValues("nearest", "rate limit",
                                        http.StatusText(result.Error.Status)).Inc()
                                // If the client provided a client_name, we want to know how many times
                                // that client_name was rate limited. This may be empty, which is fine.
                                clientName := req.Form.Get("client_name")
                                metrics.RateLimitedTotal.WithLabelValues(clientName, status.LimitType).Inc()

                                log.Printf("Rate limit (%s) exceeded for IP: %s, client: %s, UA: %s", ip,
                                        status.LimitType, clientName, ua)
                                writeResult(rw, result.Error.Status, &result)
                                return
                        }
                } else {
                        // This should never happen if Locate is deployed on AppEngine.
                        log.Println("Cannot find IP address for rate limiting.")
                }
        }

        c.handleNearestRequest(rw, req, &result, "nearest", nil)
}

// handleNearestRequest is a helper that contains the common logic for looking up
// nearest machines and generating the response. It's used by both Nearest and
// PriorityNearest handlers.
func (c *Client) handleNearestRequest(rw http.ResponseWriter, req *http.Request, result *v2.NearestResult, metricLabel string, ic *IntegrationClaims) {
        experiment, service := getExperimentAndService(req.URL.Path)

        // Look up client location.
        loc, err := c.checkClientLocation(rw, req)
        if err != nil {
                status := http.StatusServiceUnavailable
                result.Error = v2.NewError("nearest", "Failed to lookup nearest machines", status)
                writeResult(rw, result.Error.Status, result)
                metrics.RequestsTotal.WithLabelValues(metricLabel, "client location",
                        http.StatusText(result.Error.Status)).Inc()
                return
        }

        // Parse client location.
        lat, errLat := strconv.ParseFloat(loc.Latitude, 64)
        lon, errLon := strconv.ParseFloat(loc.Longitude, 64)
        if errLat != nil || errLon != nil {
                result.Error = v2.NewError("client", errFailedToLookupClient.Error(), http.StatusInternalServerError)
                writeResult(rw, result.Error.Status, result)
                metrics.RequestsTotal.WithLabelValues(metricLabel, "parse client location",
                        http.StatusText(result.Error.Status)).Inc()
                return
        }

        // Find the nearest targets using the client parameters.
        q := req.URL.Query()
        t := q.Get("machine-type")
        country := req.Header.Get("X-AppEngine-Country")
        sites := q["site"]
        org := q.Get("org")
        strict := false
        if qsStrict, err := strconv.ParseBool(q.Get("strict")); err == nil {
                strict = qsStrict
        }
        // If strict, override the country from the AppEngine header with the one in
        // the querystring.
        if strict {
                country = q.Get("country")
        }
        opts := &heartbeat.NearestOptions{Type: t, Country: country, Sites: sites, Org: org, Strict: strict}
        targetInfo, err := c.LocatorV2.Nearest(service, lat, lon, opts)
        if err != nil {
                result.Error = v2.NewError("nearest", "Failed to lookup nearest machines", http.StatusInternalServerError)
                writeResult(rw, result.Error.Status, result)
                metrics.RequestsTotal.WithLabelValues(metricLabel, "server location",
                        http.StatusText(result.Error.Status)).Inc()
                return
        }

        pOpts := paramOpts{
                raw:       req.Form,
                version:   "v2",
                ranks:     targetInfo.Ranks,
                svcParams: static.ServiceParams,
        }
        // Populate target URLs and write out response.
        c.populateURLs(targetInfo.Targets, targetInfo.URLs, experiment, pOpts, ic)
        result.Results = targetInfo.Targets
        writeResult(rw, http.StatusOK, result)
        metrics.RequestsTotal.WithLabelValues(metricLabel, "success", http.StatusText(http.StatusOK)).Inc()
}

// PriorityNearest handles requests to /v2/priority/nearest with tier-based rate limiting.
// It requires a valid integration JWT with an int_id claim. Requests without valid
// credentials receive a 401 Unauthorized response.
func (c *Client) PriorityNearest(rw http.ResponseWriter, req *http.Request) {
        result := &v2.NearestResult{}
        setHeaders(rw)

        // Handle CORS preflight requests. Use 200 instead of 204 for broader browser
        // compatibility (some Firefox versions reject 204 for CORS).
        //
        // See https://stackoverflow.com/a/46028619.
        if req.Method == http.MethodOptions {
                rw.WriteHeader(http.StatusOK)
                return
        }

        if err := req.ParseForm(); err != nil {
                log.Debugf("ParseForm error: %v", err)
                result.Error = v2.NewError("server", "Failed to parse form", http.StatusInternalServerError)
                writeResult(rw, result.Error.Status, result)
                metrics.RequestsTotal.WithLabelValues("priority_nearest", "server",
                        http.StatusText(result.Error.Status)).Inc()
                return
        }

        // Extract JWT claims from thee X-Endpoint-API-UserInfo header
        claims, err := c.extractJWTClaims(req)
        if err != nil {
                log.Debugf("Failed to extract JWT claims for priority endpoint: %v", err)
                result.Error = v2.NewError("auth", "Valid JWT token required", http.StatusUnauthorized)
                writeResult(rw, result.Error.Status, result)
                metrics.RequestsTotal.WithLabelValues("priority_nearest", "auth", http.StatusText(http.StatusUnauthorized)).Inc()
                return
        }

        // Extract integration ID claim
        intID, err := extractIntegrationID(claims)
        if err != nil {
                log.Debugf("Failed to extract int_id claim for priority endpoint: %v", err)
                result.Error = v2.NewError("auth", "Valid int_id claim required", http.StatusUnauthorized)
                writeResult(rw, result.Error.Status, result)
                metrics.RequestsTotal.WithLabelValues("priority_nearest", "auth", http.StatusText(http.StatusUnauthorized)).Inc()
                return
        }

        // Extract tier claim, defaulting to 0 if not present.
        // JSON numbers always unmarshal to float64 in Go.
        tier := 0
        if tierClaim, ok := claims["tier"]; ok {
                if v, ok := tierClaim.(float64); ok {
                        tier = int(v)
                } else {
                        log.Debugf("Unexpected tier claim type for int_id %s: %T, using default tier 0", intID, tierClaim)
                }
        }

        // Look up tier configuration
        tierConfig, ok := c.tierLimits[tier]
        if !ok {
                log.Debugf("Tier %d not configured for int_id %s, using tier 0", tier, intID)
                tier = 0
                tierConfig, ok = c.tierLimits[0]
                if !ok {
                        // Tier 0 not configured - this is a server misconfiguration
                        log.Errorf("Tier 0 not configured, cannot serve priority requests")
                        result.Error = v2.NewError("server", "Service misconfigured", http.StatusInternalServerError)
                        writeResult(rw, result.Error.Status, result)
                        metrics.RequestsTotal.WithLabelValues("priority_nearest", "config", http.StatusText(http.StatusInternalServerError)).Inc()
                        return
                }
        }

        // Apply tier-based rate limiting
        if c.ipLimiter != nil {
                ip := getRemoteAddr(req)
                if ip != "" {
                        status, err := c.ipLimiter.IsLimitedWithTier(intID, ip, tierConfig)
                        if err != nil {
                                // Log error but don't block request (fail open).
                                log.Errorf("Tier rate limiter error for int_id %s, tier %d: %v", intID, tier, err)
                        } else if status.IsLimited {
                                // Rate limited - block the request
                                result.Error = v2.NewError("client", tooManyRequests, http.StatusTooManyRequests)
                                metrics.RequestsTotal.WithLabelValues("priority_nearest", "rate limit",
                                        http.StatusText(result.Error.Status)).Inc()

                                clientName := req.Form.Get("client_name")
                                metrics.RateLimitedTotal.WithLabelValues(clientName, status.LimitType).Inc()

                                log.Debugf("Tier rate limit (%s) exceeded for int_id: %s, tier: %d, IP: %s, client: %s",
                                        status.LimitType, intID, tier, ip, clientName)
                                writeResult(rw, result.Error.Status, result)
                                return
                        }
                } else {
                        // This should never happen if Locate is deployed on AppEngine.
                        // For local deployments, RemoteAddr is assumed to always be there.
                        log.Warnf("Cannot find IP address for tier-based rate limiting.")
                }
        }

        // Extract key_id claim if present.
        keyID := ""
        if keyIDClaim, ok := claims["key_id"]; ok {
                if v, ok := keyIDClaim.(string); ok {
                        keyID = v
                }
        }

        // Rate limit passed - handle the nearest request
        ic := &IntegrationClaims{
                IntegrationID: intID,
                KeyID:         keyID,
        }
        c.handleNearestRequest(rw, req, result, "priority_nearest", ic)
}

// extractIntegrationID extracts the "int_id" claim from integration JWT claims.
func extractIntegrationID(claims map[string]interface{}) (string, error) {
        intIDClaim, ok := claims["int_id"]
        if !ok {
                return "", fmt.Errorf("int_id claim not found in JWT")
        }

        intID, ok := intIDClaim.(string)
        if !ok {
                return "", fmt.Errorf("int_id claim is not a string")
        }

        if intID == "" {
                return "", fmt.Errorf("int_id claim is empty")
        }

        return intID, nil
}

// Live is a minimal handler to indicate that the server is operating at all.
func (c *Client) Live(rw http.ResponseWriter, req *http.Request) {
        fmt.Fprintf(rw, "ok")
}

// Ready reports whether the server is working as expected and ready to serve requests.
func (c *Client) Ready(rw http.ResponseWriter, req *http.Request) {
        if c.LocatorV2.Ready() {
                fmt.Fprintf(rw, "ok")
        } else {
                rw.WriteHeader(http.StatusInternalServerError)
                fmt.Fprintf(rw, "not ready")
        }
}

// Registrations returns information about registered machines. There are 3
// supported query parameters:
//
// * format - defines the format of the returned JSON
// * org - limits results to only records for the given organization
// * exp - limits results to only records for the given experiment (e.g., ndt)
//
// The "org" and "exp" query parameters are currently only supported by the
// default or "machines" format.
func (c *Client) Registrations(rw http.ResponseWriter, req *http.Request) {
        var err error
        var result interface{}

        setHeaders(rw)

        q := req.URL.Query()
        format := q.Get("format")

        switch format {
        case "geo":
                result, err = siteinfo.Geo(c.LocatorV2.Instances(), q)
        default:
                result, err = siteinfo.Hosts(c.LocatorV2.Instances(), q)
        }

        if err != nil {
                v2Error := v2.NewError("siteinfo", err.Error(), http.StatusInternalServerError)
                writeResult(rw, http.StatusInternalServerError, v2Error)
                return
        }

        writeResult(rw, http.StatusOK, result)
}

// checkClientLocation looks up the client location and copies the location
// headers to the response writer.
func (c *Client) checkClientLocation(rw http.ResponseWriter, req *http.Request) (*clientgeo.Location, error) {
        // Lookup the client location using the client request.
        loc, err := c.Locate(req)
        if err != nil {
                return nil, errFailedToLookupClient
        }

        // Copy location headers to response writer.
        for key := range loc.Headers {
                rw.Header().Set(key, loc.Headers.Get(key))
        }

        return loc, nil
}

// populateURLs populates each set of URLs using the target configuration.
func (c *Client) populateURLs(targets []v2.Target, ports static.Ports, exp string, pOpts paramOpts, ic *IntegrationClaims) {
        for i, target := range targets {
                tkn := c.getAccessToken(target.Machine, exp, ic)
                params := c.extraParams(target.Machine, i, pOpts)
                targets[i].URLs = c.getURLs(ports, target.Hostname, tkn, params)
        }
}

// getAccessToken allocates a new access token using the given machine name as
// the intended audience and the subject as the target service.
func (c *Client) getAccessToken(machine, subject string, ic *IntegrationClaims) string {
        // Create the token. The same access token is reused for every URL of a
        // target port.
        // A uuid is added to the claims so that each new token is unique.
        cl := jwt.Claims{
                Issuer:   static.IssuerLocate,
                Subject:  subject,
                Audience: jwt.Audience{machine},
                Expiry:   jwt.NewNumericDate(time.Now().Add(time.Minute)),
                ID:       uuid.NewString(),
        }
        var t string
        var err error
        if ic != nil && ic.IntegrationID != "" {
                t, err = c.Sign(cl, *ic)
        } else {
                t, err = c.Sign(cl)
        }
        // Sign errors can only happen due to a misconfiguration of the key.
        // A good config will remain good.
        rtx.PanicOnError(err, "signing claims has failed")
        return t
}

// getURLs creates URLs for the named experiment, running on the named machine
// for each given port. Every URL will include an `access_token=` parameter,
// authorizing the measurement.
func (c *Client) getURLs(ports static.Ports, hostname, token string, extra url.Values) map[string]string {
        urls := map[string]string{}
        // For each port config, prepare the target url with access_token and
        // complete host field.
        for _, target := range ports {
                name := target.String()
                params := url.Values{}
                params.Set("access_token", token)
                for key := range extra {
                        // note: we only use the first value.
                        params.Set(key, extra.Get(key))
                }
                target.RawQuery = params.Encode()

                host := &bytes.Buffer{}
                err := c.targetTmpl.Execute(host, map[string]string{
                        "Hostname": hostname,
                        "Ports":    target.Host, // from URL template, so typically just the ":port".
                })
                rtx.PanicOnError(err, "bad template evaluation")
                target.Host = host.String()
                urls[name] = target.String()
        }
        return urls
}

// limitRequest determines whether a client request should be rate-limited.
func (c *Client) limitRequest(now time.Time, req *http.Request) bool {
        agent := req.Header.Get("User-Agent")
        l, ok := c.agentLimits[agent]
        if !ok {
                // No limit defined for user agent.
                return false
        }
        return l.IsLimited(now)
}

// setHeaders sets the response headers for "nearest" requests and
// other similar requests where we need:
//
// 1. Content-Type equal to application/json
//
// 2. Cache-Control equal to no-store
//
// 3. CORS headers
func setHeaders(rw http.ResponseWriter) {
        // Set the content-type header to imply we will return JSON
        rw.Header().Set("Content-Type", "application/json")

        // Set CORS policy to allow third-party websites to use returned resources.
        rw.Header().Set("Access-Control-Allow-Origin", "*")
        rw.Header().Set("Access-Control-Allow-Methods", "GET, OPTIONS")
        rw.Header().Set("Access-Control-Allow-Headers", "Authorization")

        // Prevent caching of result.
        // See also: https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Cache-Control
        rw.Header().Set("Cache-Control", "no-store")
}

// writeResult marshals the result and writes the result to the response writer.
func writeResult(rw http.ResponseWriter, status int, result interface{}) {
        b, err := json.MarshalIndent(result, "", "  ")
        // Errors are only possible when marshalling incompatible types, like functions.
        rtx.PanicOnError(err, "Failed to format result")
        rw.WriteHeader(status)
        rw.Write(b)
}

// getExperimentAndService takes an http request path and extracts the last two
// fields. For correct requests (e.g. "/v2/nearest/ndt/ndt5"), this will be the
// experiment name (e.g. "ndt") and the datatype (e.g. "ndt5").
func getExperimentAndService(p string) (string, string) {
        datatype := path.Base(p)
        experiment := path.Base(path.Dir(p))
        return experiment, experiment + "/" + datatype
}

// getRemoteAddr extracts the remote address from the request. When running on
// Google App Engine, the X-Forwarded-For is guaranteed to be set. The GCP load
// balancer appends the actual client IP and the load balancer IP to any existing
// X-Forwarded-For header, so the second-to-last IP is the real client address.
// When running elsewhere (including on the local machine), the RemoteAddr from
// the request is used instead.
func getRemoteAddr(req *http.Request) string {
        xff := req.Header.Get("X-Forwarded-For")
        if xff != "" {
                // Split by comma and trim spaces from each IP
                ips := strings.Split(xff, ",")
                for i := range ips {
                        ips[i] = strings.TrimSpace(ips[i])
                }

                // GCP load balancer appends: <original-header>, <client-ip>, <lb-ip>
                // The second-to-last IP is the actual client address
                if len(ips) >= 2 {
                        return ips[len(ips)-2]
                } else if len(ips) == 1 && ips[0] != "" {
                        return ips[0]
                }
        }

        // Fall back to RemoteAddr for local testing or deployments outside of GAE.
        host, _, err := net.SplitHostPort(req.RemoteAddr)
        if err != nil {
                // As a last resort, use the whole RemoteAddr.
                return strings.TrimSpace(req.RemoteAddr)
        }
        return host
}

m-lab / locate / 24666372903

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous