beads/internal/httpapi/claim.go
Julian Knutsen a7fe6aea50
feat(httpapi): PATCH /v0/beads/issues/{id} accepts claim, the wire spelling of bd update --claim (#6890)
The update operation had no way to express a claim, so an HTTP client of
`bd serve` had to refuse `bd update <id> --claim` outright. The request now
carries a `claim` boolean that passes straight through to
issueops.UpdateRequest.Claim on the same lifecycle role the direct route
uses, so the claim and the patch land in one transaction with the direct
route's semantics: same-actor re-claim is idempotent (changed: false), a
foreign holder is 409 already_claimed and an unclaimable status 409
not_claimable, both naming `claim` and carrying {id}:claim's extension
members (shared through claimRefusal), and a refused claim writes none of
the patch.

With claim: true the patch may be empty (the claim is the whole write).
claim beside expected_assignee, expected_status or force_assignee_transfer
is refused at the edge as a 400 naming `claim` rather than surfacing the
role's ErrValidation as the generic patch 400. No capability token: per the
document's version-skew rules, member presence is probed, and an older
server refuses `claim` as unknown_parameter before any database work.

Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-authored-by: Eddie the Engineer <adopt-pr@gascity.local>
2026-09-30 15:42:37 -07:00

577 lines
25 KiB
Go

package httpapi
import (
"context"
"encoding/json"
"errors"
"fmt"
"mime"
"net/http"
"slices"
"strings"
"time"
"unicode"
"unicode/utf8"
"github.com/steveyegge/beads/internal/httpapi/apigen"
"github.com/steveyegge/beads/internal/storage"
"github.com/steveyegge/beads/internal/storage/uow"
"github.com/steveyegge/beads/internal/types"
"github.com/steveyegge/beads/issueops"
"github.com/steveyegge/beads/memoryops"
)
const (
// claimActorMember is the only member ClaimRequest carries. The schema is
// additionalProperties: false, so anything else is refused by name.
claimActorMember = "actor"
// maxActorBytes is the document's cap on `actor`. The schema's maxLength
// counts characters; this byte limit is the binding one.
maxActorBytes = 256
// maxJSONBodyBytes bounds every request body on this surface. The largest
// documented one is a handful of short members, so this is pure refusal of
// the absurd.
maxJSONBodyBytes = 1 << 20
// claimContentType is the one media type this surface accepts on a body —
// requireJSONContent enforces it for every body-carrying operation, not
// only the claim — and refusing anything else is a CSRF control, not
// pedantry: a JSON content type is not CORS-"simple", so a cross-origin
// write always triggers a preflight this server never approves. Accepting
// text/plain or a form encoding would let an attacker's page skip the
// preflight and drive a write from any browser on the host.
claimContentType = "application/json"
)
// handleClaim is v0's compare-and-set claim of a single issue for a
// caller-named actor, and the write whose posture every later one adopted.
//
// ACTOR SEMANTICS, stated because adopting this endpoint depends on them. The
// actor is caller-ASSERTED provenance for the audit trail and is NOT the
// authenticated principal, even where a bearer is required: the token a
// deployment configures is shared and surface-wide, so it names nobody and
// cannot confirm or contradict the actor a request sends. Any client that
// reaches this endpoint can claim as any name, exactly as any local process
// can pass any --actor to the CLI. The CAS is therefore a correctness fence
// against CONCURRENT claims, not an authorization boundary — it guarantees
// that two racing claimants cannot both win, and guarantees nothing about who
// either of them really is. What bounds the blast radius of that posture is
// the bind: loopback by default, and beyond loopback only with a token file
// (or the explicit --insecure-no-auth).
//
// Two things a CLI claim does that this deliberately does not: hooks do not
// fire (a user-controlled subprocess per mutation is an unbounded latency
// multiplier and an orphaned child at shutdown), and the per-command
// auto-commit machinery never runs. The only durable effect is the single
// storage commit the claim role makes inside its own transaction — which is
// exactly what a proxied CLI write does today.
//
// Everything above the role here is argument validation: the path split, the
// media type, the body shape and the actor rules. The claim itself — the CAS,
// the eligibility rules, the claim pools, the transaction retry and the
// refusal vocabulary — belongs to issueops.Claimer, reached through the
// provider's own accessor.
func (s *Server) handleClaim(w http.ResponseWriter, r *http.Request) {
// The custom-method dispatcher split the id off the segment and bounded it
// before this handler was chosen at all; see customMethodTarget.
id := r.PathValue(customMethodIDValue)
if !s.requireNoQuery(w, r) {
return
}
if !s.requireJSONContent(w, r) {
return
}
actor, ok := s.claimActor(w, r)
if !ok {
return
}
claimer, err := s.claimer(r)
if err != nil {
s.failClaim(w, r, err)
return
}
result, err := claimer.Claim(r.Context(), issueops.ClaimRequest{IssueID: id, Actor: actor})
if err != nil {
s.failClaim(w, r, err)
return
}
// `already_claimed` is the wire's name for the idempotent re-claim, which
// is exactly the case the role reports as an unchanged result.
writeJSON(w, apigen.ClaimResponse{
Issue: *result.Issue,
AlreadyClaimed: !result.Changed,
})
}
// requireJSONContent enforces the request media type. It reports whether the
// request may proceed.
//
// The refusal is 400 invalid_argument naming the header rather than 415:
// 415 is not in the v0 code vocabulary, and adding a status to that vocabulary
// for this would be permanent wire surface. `param` is documented as carrying a
// header name for exactly this kind of refusal — it is what the Host middleware
// already does.
//
// The document states this rule once, at the document level, beside the
// Host-header and unknown-query-parameter rules, because it holds for every
// body-carrying operation rather than for this one.
func (s *Server) requireJSONContent(w http.ResponseWriter, r *http.Request) bool {
got := r.Header.Get("Content-Type")
if media, _, err := mime.ParseMediaType(got); err == nil && media == claimContentType {
return true
}
// Attacker-controlled, and quoted by logValue: a cross-origin POST that
// tried to skip the preflight should leave a trace naming what it sent.
requestInfo(r.Context()).refuse(got)
s.fail(w, r, InvalidArgument("Content-Type", ReasonInvalidValue,
"this operation accepts "+claimContentType+" only"))
return false
}
// claimActor decodes the request body and validates its one member.
//
// Validation happens HERE, at the wire edge, before any database work is done.
// The domain layer refuses only actor == "" (internal/storage/domain/issue.go),
// so without this a whitespace-only or megabyte actor would be persisted to the
// assignee column AND interpolated into the storage commit message — where an
// unvalidated newline forges audit-trail lines that look like separate commits.
func (s *Server) claimActor(w http.ResponseWriter, r *http.Request) (string, bool) {
members, res := decodeJSONObjectBody(w, r)
if res != nil {
s.fail(w, r, *res)
return "", false
}
var unknown []string
for name := range members {
if name != claimActorMember {
unknown = append(unknown, name)
}
}
if len(unknown) > 0 {
// Same rule as an unknown query parameter, for the same reason, and
// named the same way: one offender, chosen deterministically so a
// client dispatching on `param` never sees it depend on map order.
offender := slices.Min(unknown)
requestInfo(r.Context()).refuse(offender)
s.fail(w, r, InvalidArgument(offender, ReasonUnknownParameter,
"this operation's request body carries `"+claimActorMember+"` and nothing else"))
return "", false
}
raw, ok := members[claimActorMember]
if !ok {
s.fail(w, r, InvalidArgument(claimActorMember, ReasonInvalidValue,
"`"+claimActorMember+"` is required"))
return "", false
}
// Through a POINTER, so that `null` reaches the type-mismatch branch:
// unmarshaling JSON null into a string is a no-op, which would have let a
// null slide down to the actor rules and be reported as "empty after
// trimming" — the right status, code, param and reason attached to prose
// that misdescribes what the client sent.
var actor *string
if err := json.Unmarshal(raw, &actor); err != nil || actor == nil {
s.fail(w, r, InvalidArgument(claimActorMember, ReasonInvalidValue,
"`"+claimActorMember+"` must be a string"))
return "", false
}
trimmed, res := validateActor(*actor)
if res != nil {
s.fail(w, r, *res)
return "", false
}
return trimmed, true
}
// decodeJSONObjectBody reads the body as a JSON object of raw members. Decoding
// the members rather than the generated struct is what makes the schema's
// additionalProperties: false enforceable by NAME: encoding/json's
// DisallowUnknownFields reports the offender only inside an error string, and
// these endpoints exist to let clients stop parsing prose.
//
// Shared with the sweep, which has the same posture and the same reason for
// it; the member vocabulary each operation accepts is its own.
func decodeJSONObjectBody(w http.ResponseWriter, r *http.Request) (map[string]json.RawMessage, *Result) {
// A body with no nameable part: `param` is documented absent on exactly
// this case and present on every other 400.
unparseable := func(detail string) *Result {
res := InvalidArgument("", ReasonInvalidValue, detail)
return &res
}
var members map[string]json.RawMessage
dec := json.NewDecoder(http.MaxBytesReader(w, r.Body, maxJSONBodyBytes))
if err := dec.Decode(&members); err != nil {
var tooLarge *http.MaxBytesError
if errors.As(err, &tooLarge) {
return nil, unparseable(fmt.Sprintf("request body is larger than %d bytes", maxJSONBodyBytes))
}
return nil, unparseable("request body must be a JSON object")
}
if members == nil {
// Valid JSON, but `null`: no members to read an actor out of.
return nil, unparseable("request body must be a JSON object")
}
if dec.More() {
return nil, unparseable("request body must be a single JSON object")
}
return members, nil
}
// validateActor applies the document's actor rules in the document's order:
// trim, then refuse an empty result, an over-long value, and any control
// character.
func validateActor(actor string) (string, *Result) {
return validateNameMember(claimActorMember, actor)
}
// validateNameMember applies the actor rules to any member that NAMES SOMEONE
// and lands in a 255-character column: `actor` on the mutations, `author` on a
// comment. The rules are one statement parameterized by the member's spelling,
// rather than one statement per member — two copies would be two chances for a
// bound to drift, and the difference between them is a name.
//
// It reports the TRIMMED value, which is what reaches the role.
func validateNameMember(member, value string) (string, *Result) {
refuse := func(detail string) *Result {
res := InvalidArgument(member, ReasonInvalidValue, detail)
return &res
}
trimmed := strings.TrimSpace(value)
switch {
case trimmed == "":
return "", refuse("`" + member + "` is empty after trimming")
case len(trimmed) > maxActorBytes:
return "", refuse(fmt.Sprintf("`%s` is %d bytes; the limit is %d",
member, len(trimmed), maxActorBytes))
// The storage column holds types.MaxFieldLen (255) CHARACTERS, one fewer
// than the document's 256-byte cap allows. Refusing that one-character
// window here rather than letting it through keeps a documented-looking
// value from becoming a 500 from the assignee column, and keeps the check
// keyed on storage's own constant instead of a second copy of the number.
case types.CheckFieldLen(member, trimmed) != nil:
return "", refuse(fmt.Sprintf("`%s` is %d characters; storage holds at most %d",
member, utf8.RuneCountInString(trimmed), types.MaxFieldLen))
// Newline above all: the actor is interpolated into the storage commit
// message, so a multiline value would forge audit-trail lines. A comment's
// author is not, and is refused anyway — it lands in a column every renderer
// of the thread prints, where an unfiltered C1 introducer is an
// escape-sequence payload.
case strings.ContainsFunc(trimmed, isControlChar):
return "", refuse("`" + member + "` must not contain control characters")
}
return trimmed, nil
}
// isControlChar reports whether a rune is one no actor may carry: every Unicode
// control character (category Cc — C0, DEL, and the C1 block) plus the
// U+2028/U+2029 line separators.
//
// C1 is refused for the reason C0 is, not for tidiness: U+0085 is NEL, a line
// break on a VT-conformant terminal, so "alice<U+0085>bd: claim bd-9 by
// mallory" forges exactly the audit-trail line the C0 check exists to prevent
// once the actor reaches the storage commit message. U+009B is the one-byte CSI
// introducer, which makes an unfiltered actor an escape-sequence payload in
// anything that prints an assignee.
//
// The schema's `actor` pattern spells this same set, so what the document
// advertises and what the server refuses are one statement.
func isControlChar(r rune) bool {
return unicode.IsControl(r) || r == '\u2028' || r == '\u2029'
}
// failClaim answers a failed claim, adding the extension members a typed 409
// carries.
//
// The extensions come from issueops.ClaimConflictError, which the role fills
// from a query inside the transaction that lost the CAS — never from parsing
// fragments out of the sentinel's message ("already assigned to", "claimed
// by"). That substring classification is what a client adopting this endpoint
// gets to delete, and it can only delete it if the server never does it either.
func (s *Server) failClaim(w http.ResponseWriter, r *http.Request, err error) {
var conflict *issueops.ClaimConflictError
if !errors.As(err, &conflict) {
s.failErr(w, r, err)
return
}
s.fail(w, r, claimRefusal(err))
}
// claimRefusal classifies a refused claim and attaches the extension members
// the role's typed conflict reports. It is shared with the update's `claim`
// member, so a claim refused on either operation reads the same.
func claimRefusal(err error) Result {
res := ClassifyError(err)
var conflict *issueops.ClaimConflictError
if !errors.As(err, &conflict) {
return res
}
// `assignee` is documented with already_claimed only: an issue refused for
// its STATUS may well carry a stale assignee, and publishing it there would
// tell a client someone holds work they do not.
if res.Problem.Code == string(CodeAlreadyClaimed) && conflict.Assignee != "" {
res = res.WithAssignee(conflict.Assignee)
}
if conflict.Status != "" {
res = res.WithIssueStatus(string(conflict.Status))
}
return res
}
// timedProvider records how long a request spent obtaining units of work, so
// the claim's request line carries the same uow_ms every other route does.
// RunTxResult owns the unit-of-work lifecycle — that is the point of having one
// retry/commit implementation — so the measurement wraps the provider rather
// than the acquisition. Retry attempts ACCUMULATE: two attempts really did
// spend two acquisitions, and a claim that retried is exactly the case where
// the number matters.
type timedProvider struct {
inner uow.UnitOfWorkProvider
rec *reqInfo
}
// timedProvider carries the capability accessors, so a handler asks the
// provider it holds for the role — the same two-step a CLI command performs on
// a store — instead of reaching past it to a constructor.
var (
_ uow.IssueReaderSource = timedProvider{}
_ uow.IssueClaimerSource = timedProvider{}
_ uow.BatchCloserSource = timedProvider{}
_ uow.ReadyClaimerSource = timedProvider{}
_ uow.ReleaserSource = timedProvider{}
_ uow.IssueLifecycleSource = timedProvider{}
_ uow.WorkspaceConfigSource = timedProvider{}
_ uow.StatsReporterSource = timedProvider{}
_ uow.CycleDetectorSource = timedProvider{}
_ uow.EdgeReaderSource = timedProvider{}
_ uow.GraphCounterSource = timedProvider{}
_ uow.RelationsSource = timedProvider{}
_ uow.CommenterSource = timedProvider{}
_ uow.BlockingAnnotatorSource = timedProvider{}
_ uow.TreeWalkerSource = timedProvider{}
_ uow.ReadyCounterSource = timedProvider{}
_ uow.CounterSource = timedProvider{}
_ uow.QuerierSource = timedProvider{}
_ uow.SweeperSource = timedProvider{}
_ uow.DeleterSource = timedProvider{}
_ uow.BatchCreatorSource = timedProvider{}
_ uow.DependencyEditorSource = timedProvider{}
_ uow.MetadataCASSource = timedProvider{}
_ uow.BatchApplierSource = timedProvider{}
_ uow.MemoriesSource = timedProvider{}
_ uow.EventsJournalCursorSource = timedProvider{}
)
// IssueReader builds the reader OVER THIS WRAPPER rather than delegating to the
// wrapped provider's own accessor. Every accessor on timedProvider does this,
// and this type is the ONE place in the codebase where a constructor is the
// right call rather than an accessor: the whole purpose of the wrapper is
// that every unit of work the reader opens goes through NewUOW below and lands
// in this request's uow_ms. `p.inner.IssueReader()` would return a reader
// bound to the untimed provider and the measurement would silently read zero.
//
// THAT IS NOT A HYPOTHETICAL, which is why it now has a test. This decorator's
// layer is on NewUOW, so only a reader holding this wrapper can reach it —
// unlike telemetry's storage accessor, whose layer is on the RESULT and which
// therefore recurses and wraps. A reviewer reading the two side by side
// proposed making this one recurse for symmetry; the whole internal/httpapi
// suite passes with that change and every read route logs uow_ms=0.000
// forever. TestAReadRouteTimesTheUnitsOfWorkItsReaderOpens is the assertion
// that fails instead.
//
// That pin is per-route and there are thirteen accessors here, so
// TestEveryTimedProviderAccessorBindsToTheWrapper covers the rest structurally.
//
// The cost is that a provider whose own accessor decorated its reader would be
// bypassed here. There is one provider (doltSQLProvider) and its accessor is
// this same construction, so nothing is bypassed today — but if a decorating
// provider ever appears, this is the line that has to grow a wrap.
func (p timedProvider) IssueReader() (issueops.Reader, error) {
return uow.NewIssueReader(p)
}
// IssueClaimer builds the claimer OVER THIS WRAPPER, for the same reason and
// with the same hazard as IssueReader above: the role's units of work must go
// through NewUOW below or every claim reports uow_ms=0.000.
// TestAClaimTimesTheUnitsOfWorkItsClaimerOpens is the assertion that fails
// instead of the recursion looking correct.
func (p timedProvider) IssueClaimer() (issueops.Claimer, error) {
return uow.NewIssueClaimer(p)
}
// BatchCloser builds the many-issue close role OVER THIS WRAPPER, for the same
// reason as the roles above. Like BatchApplier it opens one of the longest
// write units of work on this surface — up to a hundred closes plus their
// blocked-state maintenance in one transaction — so a recursion here would
// report uow_ms=0.000 for exactly the requests whose timing matters most.
func (p timedProvider) BatchCloser() (issueops.BatchCloser, error) {
return uow.NewBatchCloser(p)
}
// ReadyClaimer builds the take-ready-work role OVER THIS WRAPPER, for the same
// reason and with the same hazard as IssueClaimer: it opens a write unit of
// work per call, and its scan is the longest read on this surface — it walks
// the whole ready order past rows other agents took — so a recursion here would
// report uow_ms=0.000 for exactly the requests whose timing matters most.
func (p timedProvider) ReadyClaimer() (issueops.ReadyClaimer, error) {
return uow.NewReadyClaimer(p)
}
// Releaser builds the claim-release role OVER THIS WRAPPER, for the same reason
// and with the same hazard as IssueClaimer: a release opens a write transaction
// per call, so a recursion here would report uow_ms=0.000 for every one of
// them.
func (p timedProvider) Releaser() (issueops.Releaser, error) {
return uow.NewReleaser(p)
}
// IssueLifecycle builds the guarded-mutation role OVER THIS WRAPPER, for the
// same reason and with the same hazard as IssueClaimer: this role opens the
// longest write transactions on the surface, so a claimer-style recursion here
// would report uow_ms=0.000 for exactly the requests whose timing matters most.
func (p timedProvider) IssueLifecycle() (issueops.Lifecycle, error) {
return uow.NewIssueOperations(p)
}
// WorkspaceConfig builds the settings role OVER THIS WRAPPER, for the same
// reason and with the same hazard as the two above.
func (p timedProvider) WorkspaceConfig() (issueops.WorkspaceConfig, error) {
return uow.NewWorkspaceConfig(p)
}
// StatsReporter builds the summary role OVER THIS WRAPPER, for the same reason
// and with the same hazard as IssueReader.
func (p timedProvider) StatsReporter() (issueops.StatsReporter, error) {
return uow.NewStatsReporter(p)
}
// CycleDetector builds the detector OVER THIS WRAPPER, for the same reason and
// with the same hazard as IssueReader.
func (p timedProvider) CycleDetector() (issueops.CycleDetector, error) {
return uow.NewCycleDetector(p)
}
// EdgeReader builds the stored-edge reader OVER THIS WRAPPER, for the same
// reason and with the same hazard as IssueReader.
func (p timedProvider) EdgeReader() (issueops.EdgeReader, error) {
return uow.NewEdgeReader(p)
}
// GraphCounter builds the edge-count role OVER THIS WRAPPER, for the same
// reason and with the same hazard as IssueReader.
func (p timedProvider) GraphCounter() (issueops.GraphCounter, error) {
return uow.NewGraphCounter(p)
}
// IssueRelations builds the single-anchor neighbor role OVER THIS WRAPPER, for
// the same reason and with the same hazard as IssueReader. It is the one
// accessor here whose name is not the role's: the seam spells it IssueRelations
// on both the store and the provider, and this type implements the seam.
func (p timedProvider) IssueRelations() (issueops.Relations, error) {
return uow.NewIssueRelations(p)
}
// Commenter builds the add-comment role OVER THIS WRAPPER, for the same reason
// and with the same hazard as IssueReader.
func (p timedProvider) Commenter() (issueops.Commenter, error) {
return uow.NewCommenter(p)
}
// BlockingAnnotator builds the blocking-decoration role OVER THIS WRAPPER, for
// the same reason and with the same hazard as IssueReader.
func (p timedProvider) BlockingAnnotator() (issueops.BlockingAnnotator, error) {
return uow.NewBlockingAnnotator(p)
}
// TreeWalker builds the dependency-tree walker OVER THIS WRAPPER, for the same
// reason and with the same hazard as IssueReader.
func (p timedProvider) TreeWalker() (issueops.TreeWalker, error) {
return uow.NewTreeWalker(p)
}
// ReadyCounter builds the ready counter OVER THIS WRAPPER, for the same reason
// and with the same hazard as IssueReader.
func (p timedProvider) ReadyCounter() (issueops.ReadyCounter, error) {
return uow.NewReadyCounter(p)
}
// Counter builds the issue counter OVER THIS WRAPPER, for ReadyCounter's reason.
func (p timedProvider) Counter() (issueops.Counter, error) {
return uow.NewCounter(p)
}
// Querier builds the boolean-query role OVER THIS WRAPPER, for the same reason
// and with the same hazard as IssueReader.
func (p timedProvider) Querier() (issueops.Querier, error) {
return uow.NewQuerier(p)
}
// Sweeper builds the sweeper OVER THIS WRAPPER, for the same reason and with
// the same hazard as IssueReader.
func (p timedProvider) Sweeper() (issueops.Sweeper, error) {
return uow.NewSweeper(p)
}
// Deleter builds the deleter OVER THIS WRAPPER, for the same reason and with
// the same hazard as IssueReader.
func (p timedProvider) Deleter() (issueops.Deleter, error) {
return uow.NewDeleter(p)
}
// BatchCreator builds the batch creator OVER THIS WRAPPER, for the same reason
// as the roles above. It is the one role here that opens a WRITE unit of work
// per call.
func (p timedProvider) BatchCreator() (issueops.BatchCreator, error) {
return uow.NewBatchCreator(p)
}
// DependencyEditor builds the graph's write role OVER THIS WRAPPER, for the
// same reason as the roles above. Like BatchCreator and the lifecycle it opens
// a WRITE unit of work per call.
func (p timedProvider) DependencyEditor() (issueops.DependencyEditor, error) {
return uow.NewDependencyEditor(p)
}
// MetadataCAS builds the conditional metadata write OVER THIS WRAPPER, for the
// same reason and with the same hazard as IssueReader.
func (p timedProvider) MetadataCAS() (issueops.MetadataCAS, error) {
return uow.NewMetadataCAS(p)
}
// BatchApplier builds the ordered-plan write role OVER THIS WRAPPER, for the
// same reason as the roles above. It opens the LONGEST write unit of work on
// this surface — up to a hundred items in one transaction — so a recursion here
// would report uow_ms=0.000 for exactly the requests whose timing matters most.
func (p timedProvider) BatchApplier() (issueops.BatchApplier, error) {
return uow.NewBatchApplier(p)
}
// Memories builds the persistent-memory role OVER THIS WRAPPER, for the same
// reason and with the same hazard as IssueReader. It is the one accessor here
// whose role is not an issueops role; the binding rule is the same.
func (p timedProvider) Memories() (memoryops.Memories, error) {
return uow.NewMemories(p)
}
func (p timedProvider) EventsJournalCursor() (storage.EventsJournalCursor, error) {
return uow.NewEventsJournalCursor(p)
}
func (p timedProvider) NewUOW(ctx context.Context) (uow.UnitOfWork, error) {
start := time.Now()
uw, err := p.inner.NewUOW(ctx)
if p.rec != nil {
p.rec.uowWait += time.Since(start)
}
return uw, err
}
// Close satisfies the provider interface and is never called: RunTxResult
// closes units of work, not providers. It deliberately does NOT reach the
// wrapped provider — closing the process-wide pool from inside one request
// would take the server down with it.
func (p timedProvider) Close(context.Context) error { return nil }