The idempotency assertion compared counter deltas between the two runs,
but per-run counters are reset at the start of every run, so the second
run's row already showed that run alone — the delta was negative and the
script failed before reaching anything else.
With that fixed, a second account of 3000 messages exercises the new
stop path end to end: pause mid-folder, assert the task and the account
settle into paused, resume, and assert the resumed run covers every
message while skipping the ones the paused stretch had already copied.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
A live run could only be stopped one account at a time, and stopping it
at all meant losing the queue: the accounts that had not started yet
stayed idle with no record that they were meant to run.
A run now carries a handle holding the context that stops every account
under it plus the reason it was stopped. Pause and cancel take the same
path and differ only in the status left behind — paused accounts are
what Resume re-runs, and the migration journal makes each one continue
where it stopped instead of re-copying. Accounts still queued when the
stop lands get the same status as the interrupted ones, so the whole
remainder is resumable after a pause and cancelled after a cancel.
Database writes keep using the uncancellable context, so statuses and
counters survive the stop.
The scheduler skips paused tasks: auto-starting a full run would defeat
the pause. An operator stopping a run no longer trips the schedule
breaker either — that is for failures, not for intent.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
An imported account with a wrong password could only be deleted and
re-added, which loses its folder mapping and its migration journal.
Clicking either FAIL badge now opens a dialog for both logins and both
passwords.
Passwords are never sent to the browser, so the password fields start
empty and an empty field keeps the stored ciphertext — one side can be
corrected without retyping the other. Saving resets both test verdicts
to unknown: they described the previous credentials, and the run gate
requires a passing test on both sides, so the account cannot start on an
unverified password.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Mapping the Exchange/Kerio special folders onto mailcow's by hand is
repetitive work that scales with the number of accounts. "By default"
maps Deleted Items to Trash, Junk E-mail to Junk, Sent Items to Sent and
unchecks Public Folders, which mailcow has no counterpart for.
The destination select only offered the source folder as a name to
create, so a target missing on the destination could not be selected at
all. It now also offers the current selection, and marks any name absent
from the destination as "(create)".
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Only the plain four-column format had a downloadable example, leaving
the Kerio route undocumented in the UI. Each import button now carries
its own sample link underneath: the plain comma-separated layout and a
Kerio export with the Name;FullName;Description;Enable header, a
disabled row included to show what the import skips.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The endpoints screen could only create and edit servers, so a mistyped
or retired endpoint stayed in the list forever.
Tasks reference endpoints without ON DELETE CASCADE, so a referenced
endpoint is refused with 409 and a count of the tasks using it rather
than cascading away migration history. The foreign-key violation is
mapped to the same status to cover a task created between check and
delete.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
pg_isready without -d connects to a database named after the user, but
the database is imapcopier, so every probe logged a FATAL and the
healthcheck only passed because pg_isready treats "server rejects the
connection" as reachable.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Introduce Client wrapper with socket deadline support
Add reconnection logic for body read timeouts
Implement test cases for underflow scenarios
Update orchestrator to handle reconnections
Add OnActivity callback to CopyDeps to prevent stall timeouts during large message transfers
Implement touchReader and touchWriter wrappers to call OnActivity during FETCH and APPEND operations
Add slow message logging to identify performance bottlenecks
Add test case to verify activity reporting during message transfers
Clean up orchestrator account reset code formatting
Improve account display by showing source/destination in a stacked format
when they differ, and add fixed sizing for live progress rows to prevent
resizing during scans. Also update the table header to use a single
"Account" column.
Accounts finishing done_with_errors showed only a count and a single
last_error. This adds a modal listing every concrete error of the
account's most recent run.
- migration 0005: account_errors table (kind folder|message|account,
folder, message_ref, error, created_at; ON DELETE CASCADE; indexed)
- store: AddAccountError / ClearAccountErrors / ListAccountErrors
- copy: OnError callback captures per-message error text (previously
only counted), with a "UID N: subject" reference
- orchestrator: clear errors at run start; persist folder/message/
account errors; cap 500 rows/account/run with a suppressed-note row
- api: GET /api/tasks/{id}/accounts/{accountId}/errors
- web: AccountErrorsModal, clickable ERRORS count, api + styles
Verified: migration applies on Postgres 18; store add/list/clear and
cascade tests pass against real pg; backend build/vet/test green; web
tsc+vite build and oxlint clean.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01U9Eq4JtWjyNTv5qat3B3mM
Introduce idleConn wrapper to prevent wedged connections
Add test for silent server timeout behavior
Implement proper TLS handshake and greeting handling
Set explicit dial and read timeouts
A 30s in-process scheduler auto-runs tasks on a recurring interval
(1/3/6/12/24h). Interval is measured from the last run's completion; a task
already running is never re-triggered. Enabling a schedule requires all
accounts tested OK. A breaker disables the schedule and flags the task broken
(red badge in list + detail) when a scheduled run errors or panics. Next-run
time (browser-local) and a run-log modal (per-run status/totals) are shown.
Migration 0004 adds tasks.schedule_interval_seconds/schedule_anchor/broken and
runs.trigger; task delete cascades runs as before.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Move folder config from task-level to per-account (accounts.folder_mapping +
excluded_folders, migrating existing task mapping). Add per-account folders
button that re-probes an account and opens the mapping modal with sync
checkboxes to exclude folders. Orchestrator builds each account's plan via
planFolders honoring per-account mapping/exclusions.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Error consistency: folder-level failures were counted only toward the task's
done_with_errors status, not the account's error_count, so a task showed
DONE_WITH_ERRORS while its only account showed 0 errors / DONE. Now folder
errors increment the account counter and the account status becomes
done_with_errors when errs>0.
Visibility: persist accounts.last_error (migration 0002) so the failing folder
/ login error survives a page reload (shown red under the source login);
cleared at the start of each run.
Modal reset: the folder-mapping modal kept its selections across opens, so
adding a second account showed the first account's mapping. It now mounts
fresh per add (conditional render + key), reflecting the newly-probed folders.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MMHQTtnQtQqL8muAXHr9kd
ADD now probes both connections, lists folders on each side, and opens a
mapping modal to route source->destination folders (e.g. Спам -> Spam) so we
append into the existing folder instead of creating a duplicate.
- store: SetTaskFolderMapping (+ round-trip test)
- httpapi: POST /tasks/{id}/probe (test both, return folder lists),
PUT /tasks/{id}/folder-mapping
- web: FolderMappingModal (reuses Modal, size=lg), submitAccount probes then
opens the modal; confirm creates the account and saves the task mapping
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MMHQTtnQtQqL8muAXHr9kd
CopyFolder now streams envelope metadata via Next() in a first pass (dedup +
queue new messages), then streams bodies for new ones in a second pass —
no more blocking Collect of the whole folder with zero feedback, and memory
stays flat (only new-message meta is held).
- imapx: two-pass streaming CopyFolder + CopyDeps.OnScan(scanned,total)
- orchestrator: throttled 'scan' events during the metadata pass
- web: per-account 'scanning folder: X/N' line under the progress bar;
scan events kept out of the log to avoid flooding
Verified on greenmail: idempotency and internal-date preservation still hold.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MMHQTtnQtQqL8muAXHr9kd
Before copying, EXAMINE every folder to sum the account's total message count
and emit a 'plan' event; progress events now carry account_total so the UI
shows a real overall bar, percent and ETA (not just per-folder).
- imapx.FolderMessageCount: read-only count of a folder
- orchestrator: plan pass + grandTotal, plan event, account_total in progress
- web: live progress keyed on account total; PLAN log line; overall bar/ETA
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MMHQTtnQtQqL8muAXHr9kd
- orchestrator: progress events now carry account-level cumulative copied/
skipped plus current folder done/total, throttled to ~3/sec per account
- web: RUN CONTROL counters and account copied/skipped read live WS values
(DB only advances per folder, so the summary lagged); new Progress column
shows a bar, percent, avg messages/sec and folder ETA while running
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MMHQTtnQtQqL8muAXHr9kd
The run-cancel registry is in-memory; a container restart mid-run leaves
accounts/tasks persisted as 'running' with no goroutine, wedging cancel
(not-in-map -> 409) and blocking remove/re-run.
- startup: ResetRunningOnStartup clears stale 'running' -> 'idle' on boot
- cancel handler: when no live goroutine, ClearStuckAccount + ReconcileTaskStatus
reset the stuck account (and its task) instead of returning 409
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MMHQTtnQtQqL8muAXHr9kd
- orchestrator: per-account cancellable context registry + CancelAccount;
on cancel, close IMAP connections to unblock in-flight FETCH; account ends
in 'cancelled' status with a cancelled event
- imapx: CopyDeps.OnFolder callback fires after EXAMINE with the folder's
message count (before the long fetch) for visibility
- httpapi: POST /tasks/{id}/accounts/{accountId}/cancel
- web: per-row cancel button while running, folder event shows N messages,
cancelled/done_with_errors status badges
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MMHQTtnQtQqL8muAXHr9kd
Pasted app passwords (e.g. mail.ru) often carry a trailing space/newline that
the IMAP server rejects; the CSV path already trims, so make manual add match.
Source and destination passwords remain fully independent (src_pass_enc /
dst_pass_enc), verified end-to-end — this only strips surrounding whitespace.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MMHQTtnQtQqL8muAXHr9kd
Modal component: portal, ESC to close, Tab focus-trap, focus-on-open
(prefers [data-modal-autofocus]), focus restore, overlay click, scroll lock.
ConfirmProvider exposes useConfirm(): async confirm({...}) as a drop-in for
window.confirm; Enter confirms, ESC cancels. Task/account deletes now use it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MMHQTtnQtQqL8muAXHr9kd
Root cause: store List* methods used 'var out []T' which stays nil on empty
result sets and serializes to JSON null; the SPA then crashed on .length/.map
(e.g. endpoints.length on the Tasks page right after deploy). Return []T{}
at the source; coerce null->[] on the frontend load sites as defense-in-depth.
Regression test asserts List* are non-nil when empty.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MMHQTtnQtQqL8muAXHr9kd
Go's encoding/json does not bridge snake_case <-> PascalCase field names,
so store.Endpoint, store.Task and the anonymous request bodies in
accounts.go/auth.go were silently decoding empty/zero values from the
frontend's snake_case JSON contract (tls_mode, role_label,
src_endpoint_id, dst_endpoint_id, src_login/pass, dst_login/pass).
Adds explicit json tags; DB layer is unaffected since pgx binds by
positional params, not struct-tag reflection.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MMHQTtnQtQqL8muAXHr9kd
Vite + React 19 + TS console-style operator UI: hash-routed Login,
Endpoints, Tasks, and TaskDetail (realtime accounts table over /ws,
Run gated on all accounts testing ok on both sides).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MMHQTtnQtQqL8muAXHr9kd
vasyanskandClaude Opus 5