Ressourcen · 37
Sichern Sie eine GraphQL-API: Autorisierung, Abfragekosten und Beweise
Erzwingen Sie Objekt- und Feldrechte, binden Sie Serverarbeit und testen Sie Verweigerungspfade.
Aktualisiert · 7 min
Wozu dieser Leitfaden beiträgt
- Ordnen Sie das verfügbar gemachte Schema zu
- Ausübungsberechtigungen
- Begrenzte Abfragekosten
- Ablehnungen überprüfen
Kurzcheck
- Kann ein Konto das Objekt eines anderen Kontos lesen?
- Folgt ein sensibles Feld der Zugriffsentscheidung seines übergeordneten Feldes?
- Was kosten Tiefe, Aliase und Paginierung zusammen?
- Welche Details dringen durch Fehler durch?
- Wer beobachtet Verleugnungen und Beschimpfungen?
Schritt-für-Schritt-Anleitung
- 1
Inventurarbeiten
Listentypen, Felder, Mutationen, Rollen, Objektbesitzer und sensible Daten. Beziehen Sie aktuelle Clientabfragen und selten verwendete Pfade ein.
Lieferinhalt: Operation, Rolle, Objekt und Feldmatrix.
- 2
Testen Sie jede Zugriffsentscheidung
Erstellen Sie unterschiedliche Testkonten und tauschen Sie Objekt-IDs aus, einschließlich verschachtelter Objekte. Resolver müssen die Autorisierung durchsetzen, bevor sie Daten zurückgeben.
Lieferbar: reproduzierbare zugelassene und abgelehnte Fälle.
- 3
Gebundene Serverarbeit
Messen Sie die tatsächlichen Kosten von Feldern und Sammlungen. Testen Sie Tiefe, Breite, Aliase, Paginierung, Stapelung und Zeit zusammen und legen Sie dann geeignete Grenzen fest.
Liefergegenstand: Kostenpolitik und gegnerischer Abfragesatz.
- 4
Reduzieren Sie unnötige Belichtung
Entscheiden Sie sich für eine Introspektion des Kontexts, blenden Sie interne Fehlerdetails aus und entfernen Sie Felder, die mehr verraten, als Kunden benötigen.
Liefergegenstand: Überprüfte öffentliche Konfiguration und Fehlerantworten.
- 5
Beobachten Sie mit Zurückhaltung
Protokollieren Sie nützliche technische Identität, Betrieb, Kosten, Ablehnung und Korrelation und schließen Sie dabei Geheimnisse und sensible Argumente aus. Weisen Sie den Eigentümer der Warnung zu.
Liefergegenstand: Dashboard und Untersuchungsverfahren.
- 6
Wiederholung nach Änderung
Wenn sich ein Feld, eine Rolle oder ein Mandant ändert, wiederholen Sie Berechtigungs- und Lastfälle. Überprüfen Sie alte Abfragen, die jetzt das überarbeitete Diagramm durchlaufen.
Liefergegenstand: Regressionsergebnis und Freigabeentscheidung.
Managementindikatoren
| Indikator | Was es misst | Erste Aktion |
|---|---|---|
| Zugang | Objekt- und Feldfälle wurden wie erwartet abgelehnt | Den Resolver reparieren |
| Kosten | Schwere Abfragen durch Richtlinie enthalten | Felder und Grenzen anpassen |
| Fehler | Öffentliche Antworten ohne interne Details | Belichtete Details reduzieren |
| Regression | Client-Vorgänge werden nach Änderung wiedergegeben | Blockieren Sie unbeabsichtigten Bruch |
Häufige Fallstricke
- Allein dem Endpunktzugriff vertrauen
- Tiefe begrenzen, Breite aber ignorieren
- Autorisierung verschachtelter Felder vergessen
- Argumente mit sensiblen Daten protokollieren
Häufig gestellte Fragen
Entfernt GraphQL die normale API-Autorisierung?
Nein. Jeder Objekt- und Feldzugriff erfordert weiterhin eine identitäts- und kontextbasierte Entscheidung.
Reicht es aus, die Selbstbeobachtung zu deaktivieren?
Nein. Es reduziert möglicherweise die Schema-Exposition, korrigiert jedoch nicht die Zugriffs- oder Abfragekosten.
Reicht eine Tiefenbegrenzung?
Nein. Aliase, Breite, Listen und feldspezifische Kosten spielen ebenfalls eine Rolle.
GraphQL: Fehler, Listen und Teildaten
Ausführliche Hinweise stehen in dieser Ausgabe auf Englisch zur Verfügung.
How should a client read a GraphQL response when part of the data fails? Start with the processing phase, then follow Non-Null constraints to the first position that permits null. A screen that displays success simply because transport returned can hide an incomplete record. This dossier connects schema, JSON result and interface decisions through a fictional evidence-viewing scenario.
Reference: the published September 2025 edition, reviewed on 23 October 2026. The eight rules below are a CIE editorial selection, not a conformance certification. All scenarios and sample data are fictional.
GraphQL Specification · September 2025 · Published edition · September 2025
Eight points for interpreting the result
| Point | Client consequence | Source |
|---|---|---|
Request error result | An error preventing execution yields errors and no data entry. Invalid variables follow this route even when the document parses. | GraphQL 2025 |
Execution result | After execution, data is present. Its value may be a partial object or null. With no raised error, errors must be absent rather than an empty list. | GraphQL 2025 |
errors[].message | Each error includes a developer-facing message. Give visitors useful wording without exposing internal details or sensitive identifiers. | GraphQL 2025 |
errors[].locations | When a document position can be associated with an error, locations should give line and column starting at 1. These are not data paths. | GraphQL 2025 |
errors[].path | The path uses response names, including aliases, and zero-based integer indices. It keeps the failing position even when a parent is subsequently replaced with null. | GraphQL 2025 |
List / Non-Null | Separate list nullability from item nullability. A failing non-null item nulls the list; a non-null list propagates the error further to its parent. | GraphQL 2025 |
extensions | Response and error extensions are metadata maps defined by implementations. A code in extensions is not a universal GraphQL error-code catalogue. | GraphQL 2025 |
GraphQL / HTTP | The GraphQL specification does not define transport. GraphQL over HTTP is identified here as a working draft; test its media-type negotiation separately. | HTTP draft |
Identify the phase before choosing the message
| Situation | Expected shape | Journey action |
|---|---|---|
| Request rejected before execution | Non-empty errors; data absent | Fix the operation or its variables. Keep an actionable error instead of automatically replaying an unchanged request. |
| Execution with no raised error | data present; errors absent | Display returned data. A schema-permitted null may be normal and alone does not diagnose failure. |
| Execution with partial result | Object data; non-empty errors | Separate retained fields from failing positions. Prevent confirmation that depends on unavailable evidence. |
| Propagation reaches the root | data: null; non-empty errors | Present an unavailable state for this operation. Do not silently recycle a response belonging to another case. |
GraphQL-Nullbarkeitsmodell
The model fixes a query using affaire as the alias of dossier. The case contains reference: String! and preuves, a three-item String list. statut: String! is a root sibling. Items are A-01, A-02 and A-03 before the injected failure. Choose the preuves type, dossier nullability and failing item.
query Demo {
affaire: dossier { reference preuves }
statut
}Local simulation of one item-resolution error on this fixed schema. No API call, query parser, permission test, cost calculation or HTTP verdict. Nested types, subscriptions and incremental delivery are outside the model. The JSON illustrates propagation and contains no real investigation data.
Fiktives JSON-Ergebnis
Four list types, two distinct constraints
| preuves type | Null list allowed | Null item allowed | Single item failure |
|---|---|---|---|
| [String] | Yes | Yes | One item becomes null. |
| [String!] | Yes | No | The list becomes null. |
| [String]! | No | Yes | One item becomes null; the list remains present. |
| [String!]! | No | No | Propagation reaches dossier, then the root depending on its type. |
Eight reference results, readable without JavaScript
G01 · [String] · affaire: Dossier
Only item 1 becomes null; the other two items and the case remain readable.
{
"data": {
"affaire": {
"reference": "DEMO-01",
"preuves": [
"A-01",
null,
"A-03"
]
},
"statut": "fictif"
},
"errors": [
{
"message": "Illustrative item resolution failure",
"path": [
"affaire",
"preuves",
1
]
}
]
}Acceptance check : Check the received JSON and retain errors[0].path = ["affaire", "preuves", 1], even when affaire is replaced by null.
G02 · [String!] · affaire: Dossier
The preuves list becomes null; the case reference and status remain readable.
{
"data": {
"affaire": {
"reference": "DEMO-01",
"preuves": null
},
"statut": "fictif"
},
"errors": [
{
"message": "Illustrative item resolution failure",
"path": [
"affaire",
"preuves",
1
]
}
]
}Acceptance check : Check the received JSON and retain errors[0].path = ["affaire", "preuves", 1], even when affaire is replaced by null.
G03 · [String]! · affaire: Dossier
Only item 1 becomes null; the other two items and the case remain readable.
{
"data": {
"affaire": {
"reference": "DEMO-01",
"preuves": [
"A-01",
null,
"A-03"
]
},
"statut": "fictif"
},
"errors": [
{
"message": "Illustrative item resolution failure",
"path": [
"affaire",
"preuves",
1
]
}
]
}Acceptance check : Check the received JSON and retain errors[0].path = ["affaire", "preuves", 1], even when affaire is replaced by null.
G04 · [String!]! · affaire: Dossier
The affaire case becomes null; the sibling statut field remains available.
{
"data": {
"affaire": null,
"statut": "fictif"
},
"errors": [
{
"message": "Illustrative item resolution failure",
"path": [
"affaire",
"preuves",
1
]
}
]
}Acceptance check : Check the received JSON and retain errors[0].path = ["affaire", "preuves", 1], even when affaire is replaced by null.
G05 · [String] · affaire: Dossier!
Only item 1 becomes null; the other two items and the case remain readable.
{
"data": {
"affaire": {
"reference": "DEMO-01",
"preuves": [
"A-01",
null,
"A-03"
]
},
"statut": "fictif"
},
"errors": [
{
"message": "Illustrative item resolution failure",
"path": [
"affaire",
"preuves",
1
]
}
]
}Acceptance check : Check the received JSON and retain errors[0].path = ["affaire", "preuves", 1], even when affaire is replaced by null.
G06 · [String!] · affaire: Dossier!
The preuves list becomes null; the case reference and status remain readable.
{
"data": {
"affaire": {
"reference": "DEMO-01",
"preuves": null
},
"statut": "fictif"
},
"errors": [
{
"message": "Illustrative item resolution failure",
"path": [
"affaire",
"preuves",
1
]
}
]
}Acceptance check : Check the received JSON and retain errors[0].path = ["affaire", "preuves", 1], even when affaire is replaced by null.
G07 · [String]! · affaire: Dossier!
Only item 1 becomes null; the other two items and the case remain readable.
{
"data": {
"affaire": {
"reference": "DEMO-01",
"preuves": [
"A-01",
null,
"A-03"
]
},
"statut": "fictif"
},
"errors": [
{
"message": "Illustrative item resolution failure",
"path": [
"affaire",
"preuves",
1
]
}
]
}Acceptance check : Check the received JSON and retain errors[0].path = ["affaire", "preuves", 1], even when affaire is replaced by null.
G08 · [String!]! · affaire: Dossier!
data becomes null: the chain of non-null response positions reaches the root.
{
"data": null,
"errors": [
{
"message": "Illustrative item resolution failure",
"path": [
"affaire",
"preuves",
1
]
}
]
}Acceptance check : Check the received JSON and retain errors[0].path = ["affaire", "preuves", 1], even when affaire is replaced by null.
Six business checks for a GraphQL viewing API
| CIE scenario | Evidence to collect | Decision criterion |
|---|---|---|
| G-ACC · Another tenant’s object | Two test identities and the same case ID, without production data. | The service enforces access on the object, relations and fields; possessing an ID grants no access. |
| G-ALIAS · Two aliases, one object | Two aliased selections and correlated traces in an authorized environment. | Aliases change neither access scope nor cost policy. The client links each error to the received response name. |
| G-CACHE · Account change | The sign-out journey followed by sign-in as another test account. | Cached data does not cross identity changes. Partial responses do not present an old proof as newly read. |
| G-COST · Volume and depth | Test measurements of cardinality, depth, time and service-specific quotas. | Depth alone is insufficient: bound list size, expensive fields and grouped operations. No universal threshold is invented. |
| G-ERROR · Public message | Examples of visitor-facing messages and operator-only traces. | Expose no stack, secret, table name or sensitive evidence. A non-sensitive correlation identifier supports troubleshooting. |
| G-SXO · Missing evidence blocks an action | Keyboard and screen-reader journey with a required field failing. | Announce the partial state, keep retained data readable, and explain then prevent confirmation that depends on the unavailable field. |
OWASP · GraphQL Cheat Sheet · Living guidance consulted 2026-10-23
A partial response proves neither a leak nor correct protection. Only service-side authorization tests can verify access. Link each acceptance case to the deployed schema, its version and client observations: data controls, interface resilience and transport need coordinated checks with distinct criteria.
Provenance, edition and licence
GraphQL Specification · September 2025 · reviewed 2026-10-23. CIE: original explanations, fictional scenarios and limited educational model.
GraphQL HTML SHA-256: 39fb3aeffd9b5acebb0830ddc6dcc243c1280a9f0e977897ae59c5cc165511d6
HTTP draft HTML SHA-256: 90a686f8251073b05e105697c6028288520d0c0fa6442924c0d1751cad3f14ac
GraphQL over HTTP · Current Working Draft · Working draft consulted 2026-10-23 · may change
GraphQL · Copyright and Licensing · Open Web Foundation Agreement 1.0
Offizielle Referenzen
Die Referenzen unterstützen die Methode. Passen Sie die Prüfungen an Ihren Kontext an; sie stellen keine Zertifizierung dar. Originalreferenztitel und Quelldokumente können in einer anderen Sprache verfasst sein.
Referenzen geprüft am .
