Project

General

Profile

Download (6.07 KB)

Bug #153 » README.md

README: Tolerate slow or restricted OTP mailbox sites during the NSDL login - Akshath Shettigara, 10/15/2026 04:35 PM

 

type: redmine-readme
project: Canofy Process Standardization
subject: Tolerate slow or restricted OTP mailbox sites during the NSDL login
story_key: nsdl-otp-mailbox-resilience
generated_at: 2026-10-15T16:16:51+05:30
story_type: bug
attachments:

  • file: nsdl-otp-mailbox-resilience_20261015-1616__test_transient.py
    description: Local test with a simulated slow Frappe site (test values only)

Tolerate slow or restricted OTP mailbox sites during the NSDL login

A slow answer from the Frappe site holding the OTP mailbox no longer aborts the NSDL login, and the request timeout is configurable.

1. Summary

When the OTP mailbox was moved to a different Frappe site (slc), Phase A failed in two ways. First, the API user could see a count of Communications but an empty list, because the site hides Communications of an Email Account that is not assigned to the user. After the user fixed that, a second problem appeared: the site needed 19 to 21 seconds per query (about 119,000 Communications), and the mailbox code allowed 30 seconds and treated one timeout as fatal, so a single slow answer ended the login after NSDL had already issued the OTP. Timeouts are now retried until the wait ends, and the timeout value is a setting. The user then switched the mailbox back to eapx, which answers in a second or two.

2. Requirement

# Requirement Source Status
R1 A slow Frappe answer must not abort a login whose OTP was already requested observed failure Done
R2 Real errors (permission, not found, bad configuration) must still fail at once design Done
R3 Request timeout configurable design Done
R4 Diagnose why no OTP mail was found on the new site user (2026-10-14) Done (two causes, see below)

3. Decisions

Decision Options considered Why this one Date
Treat timeouts as transient only while waiting for the mail Retry everything; retry timeouts only A 403 or 404 means misconfiguration and must not loop 2026-10-14
Baseline snapshot No retry; up to 3 tries The snapshot runs before the Login click, where a failure is still cheap 2026-10-14
Timeout default 30 s kept; 90 s in the template 30 s stays the code default; the template sets 90 s for both functions 2026-10-14
Mailbox site Keep slc; return to eapx eapx was reliable in earlier runs (user's choice) 2026-10-14

4. Logic & Workflow

click Login -> poll Communication -> answer? check for OTP : timeout or connection error -> log, keep polling until the wait ends -> end as "no OTP mail" if nothing else

  1. A timeout or connection error raises OtpMailboxTransientError, a subclass of OtpMailboxError.
  2. While waiting for the mail the error is logged and polling continues. If every attempt times out, the result is a clean "no OTP mail" rather than a crash.
  3. The baseline snapshot before the Login click is retried up to 3 times.
  4. Other errors stay fatal.

Diagnosis on 2026-10-14:

Observation Cause
HTTP 403 reading the Email Account API user lacked access; fixed by the user
Count of Communications returned, list empty Frappe's per-record filter: the Email Account was not in the API user's Email Accounts table; fixed by the user
Queries took 19 to 21 s Large Communication table (about 119,000 rows) on the slc site
pull_emails returned 417 for the API user The Lambda cannot force an import and depends on the site's own mail scheduler

5. Files & Components Changed

File / resource Type Change Purpose
shared/otp_mailbox.py Code OtpMailboxTransientError; retry while waiting; baseline retry Resilience
shared/nsdl_login.py Code Reads and validates the new setting Configuration
template.yaml SAM template NSDL_OTP_REQUEST_TIMEOUT_SECONDS: "90" on both functions Setting

6. Configuration & Deployment

Setting Allowed Default Template
NSDL_OTP_REQUEST_TIMEOUT_SECONDS 10 to 300 30 90

For a new mailbox site: the API user needs read access to Communication and Email Account, and the Email Account must be listed in that user's Email Accounts table. Deployed through the pipeline after the pull request was merged (2026-10-14).

7. Testing

Test Data / Site Expected Actual Status
Survive repeated timeouts Simulated Frappe Mail still returned Returned after 2 timeouts Pass
Timeouts for the whole wait Simulated Frappe Clean "no OTP mail" As expected Pass
Permission error Simulated 403 Fatal at once Fatal Pass
Baseline retry Simulated Frappe Survives 2 timeouts, gives up after 3 As expected Pass
Earlier OTP and no-data suites, Ruff, compile, sam validate Local Pass Pass Pass
Live run on the slc mailbox Real NSDL Login Failed before the fix (see Summary) Fail, fixed in code
Live Phase A after switching back to eapx (before this fix was deployed) Real NSDL, 2026-10-14 Login OTP mail after 60 s, accepted at request 1 without resend; login 79 s Pass
Live Phase B run, same day (this fix was not yet deployed in that run) Real NSDL Login OTP mail took 282 s to reach Frappe; the run used 14 min 40 s of the 15-minute limit Pass, with little margin

8. Known Limitations & Open Items

  • The slc site's slow Communication queries were not investigated; a missing index or archiving was suggested to its owner. Not confirmed.
  • The slc Email Account (Office 365 over IMAP) had never synced; whether it needs OAuth was not confirmed.
  • Intermittent missing OTP mails on later mornings are tracked in nsdl-eventbridge-scheduling.

9. Status

Item Value
Branch feature/AI-TBD-nsdl-otp-login
Commits e2a1c02
PR Created and merged by the user; number not confirmed
Deployed to TEST stack

10. References

Related stories: nsdl-email-otp-login, nsdl-login-otp-captcha-update, nsdl-eventbridge-scheduling.

(1-1/3)