S3 · size S · part of inbuxa for Kubernetes (tracking issue in inbuxa-charts)
Measure a webmail session sealed into a cookie (AES-GCM, key from APP_SECRET) holding real 27.1-rc tokens.
Done when
Size after sealing and base64 is under 3.5 KB, or a split-cookie design is written up
Decides the shape of K-W1
**S3** · size S · part of inbuxa for Kubernetes (tracking issue in inbuxa-charts)
Measure a webmail session sealed into a cookie (AES-GCM, key from APP_SECRET) holding real 27.1-rc tokens.
**Done when**
- Size after sealing and base64 is under 3.5 KB, or a split-cookie design is written up
- Decides the shape of K-W1
jcoffey-dev
added this to the 27.1 milestone 2026-11-18 05:14:37 +00:00
Answer, part 1 (from the code, 2026-11-17): size isn't the problem; what the cookie can't do is.
Size. Server tokens are small sealed structs (grant type, account id, issue time, expiry, credential version) with nonce and tag, around 100 bytes each base64. A sealed session (username, account, TokenSet, timestamps, remember, UA) is well under 1 KB. To confirm with a real token in the kind lab.
What a cookie-only session loses.server/src/sessions.ts already plans a stateless backend and names the problem: listForUser (the sessions list) and destroyAllForUser (sign out my other sessions) need something enumerable. The plan was to use the server's token registry, but the server has none: tokens are sealed, not stored. Revocation today is per password (credential_version), so changing a password still signs out everywhere, which is the important guarantee.
Options:
27.1: stateless cookies; in stateless mode the webmail hides the sessions list and 'sign out other sessions'; sign-out revokes its own tokens through a short-lived denylist (K-S4); password changes keep revoking everything. A grant registry in the server comes after 27.1.
27.1: build a grant registry in the server now (grant id per refresh token, list and revoke per account; also gives the console a sessions view). Bigger (L) and lands on the server's 27.1 critical path.
Recommendation: option 1. Password sessions ({u,p}) can't be revoked statelessly, so stateless mode should be OAuth-only, which is how the webmail signs in against inbuxa-server anyway.
**Answer, part 1 (from the code, 2026-11-17): size isn't the problem; what the cookie can't do is.**
**Size.** Server tokens are small sealed structs (grant type, account id, issue time, expiry, credential version) with nonce and tag, around 100 bytes each base64. A sealed session (username, account, TokenSet, timestamps, remember, UA) is well under 1 KB. To confirm with a real token in the kind lab.
**What a cookie-only session loses.** `server/src/sessions.ts` already plans a stateless backend and names the problem: `listForUser` (the sessions list) and `destroyAllForUser` (sign out my other sessions) need something enumerable. The plan was to use the server's token registry, but **the server has none**: tokens are sealed, not stored. Revocation today is per password (`credential_version`), so changing a password still signs out everywhere, which is the important guarantee.
Options:
1. **27.1:** stateless cookies; in stateless mode the webmail hides the sessions list and 'sign out other sessions'; sign-out revokes its own tokens through a short-lived denylist (K-S4); password changes keep revoking everything. A grant registry in the server comes after 27.1.
2. **27.1:** build a grant registry in the server now (grant id per refresh token, list and revoke per account; also gives the console a sessions view). Bigger (L) and lands on the server's 27.1 critical path.
Recommendation: option 1. Password sessions (`{u,p}`) can't be revoked statelessly, so stateless mode should be OAuth-only, which is how the webmail signs in against inbuxa-server anyway.
Decided 2026-11-17: option 1. Stateless cookies for 27.1; stateless mode is OAuth-only and hides the sessions list and 'sign out other sessions'; sign-out revokes its own tokens (K-S4); password changes keep revoking everything. A grant registry in the server comes after 27.1.
**Decided 2026-11-17: option 1.** Stateless cookies for 27.1; stateless mode is OAuth-only and hides the sessions list and 'sign out other sessions'; sign-out revokes its own tokens (K-S4); password changes keep revoking everything. A grant registry in the server comes after 27.1.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
S3 · size S · part of inbuxa for Kubernetes (tracking issue in inbuxa-charts)
Measure a webmail session sealed into a cookie (AES-GCM, key from APP_SECRET) holding real 27.1-rc tokens.
Done when
Answer, part 1 (from the code, 2026-11-17): size isn't the problem; what the cookie can't do is.
Size. Server tokens are small sealed structs (grant type, account id, issue time, expiry, credential version) with nonce and tag, around 100 bytes each base64. A sealed session (username, account, TokenSet, timestamps, remember, UA) is well under 1 KB. To confirm with a real token in the kind lab.
What a cookie-only session loses.
server/src/sessions.tsalready plans a stateless backend and names the problem:listForUser(the sessions list) anddestroyAllForUser(sign out my other sessions) need something enumerable. The plan was to use the server's token registry, but the server has none: tokens are sealed, not stored. Revocation today is per password (credential_version), so changing a password still signs out everywhere, which is the important guarantee.Options:
Recommendation: option 1. Password sessions (
{u,p}) can't be revoked statelessly, so stateless mode should be OAuth-only, which is how the webmail signs in against inbuxa-server anyway.Decided 2026-11-17: option 1. Stateless cookies for 27.1; stateless mode is OAuth-only and hides the sessions list and 'sign out other sessions'; sign-out revokes its own tokens (K-S4); password changes keep revoking everything. A grant registry in the server comes after 27.1.
Folded into the checklist on inbuxa/inbuxa-charts#14 to keep the issue list short. The details stay here; reopen this when work on it starts.