- Python 92.4%
- Shell 7.6%
filtermail delivers outgoing mail with one worker per destination domain and postfix waits for all domains of a message, so one bad destination (invalid TLS certificate, permanent rejection, hang) stalls whole group messages: "421 Transport busy", 600 s end-of-data timeouts, duplicates. transport-autofix reads the postfix/lmtp-filtermail/lmtp journal lines every 2 minutes, keeps 30 minutes of per-domain results and routes a domain with at least 20 failed deliveries and at most 10% successful ones through plain smtp for 24 hours (map /etc/postfix/transport_auto). Manual transport entries, the discard list and the relay's own domain are never touched. Dry-run by default, every change goes to ntfy. Tests cover parsing and the routing decision. |
||
|---|---|---|
| systemd | ||
| tests | ||
| .gitignore | ||
| chatmail-apps-mirror.py | ||
| chatmail-kuma-push.sh | ||
| chatmail-ntfy-daily.sh | ||
| chatmail-stats.py | ||
| LICENSE | ||
| README.md | ||
| transport-autofix.py | ||
chatmail-relay-tools
Small, dependency-free admin scripts for chatmail relays: count accounts and live IMAP load, send the numbers to ntfy or Uptime Kuma, and mirror the webxdc App Picker for users who cannot reach the upstream.
Everything runs on the relay itself as root and uses only what a chatmail relay
already has (python3, doveadm, coreutils, curl). Nothing here is touched by
cmdeploy run.
| Script | What it does |
|---|---|
chatmail-stats.py |
Accounts (total, active today / 7d / 30d / 90d, new 24h / 7d), IMAP sessions from Dovecot, imap processes, TCP on 993/143/443, load. Human, JSON or CSV output. |
chatmail-ntfy-daily.sh |
Once-a-day summary to an ntfy topic, with the change in account count since the previous run. |
chatmail-kuma-push.sh |
Sends numbers to Uptime Kuma Push monitors so Kuma graphs them. |
chatmail-apps-mirror.py |
Local copy of the webxdc App Picker (apps list, .xdc files, store page) at https://<relay>/apps/, refreshed by a systemd timer. |
transport-autofix.py |
Watches filtermail deliveries and, for a while, sends a destination that keeps failing (bad TLS certificate, permanent rejection, hang) through plain postfix smtp, so it no longer stalls whole group messages. Dry-run by default. |
How the numbers are counted
- Accounts: every
<mailboxes_dir>/<addr>/passwordis one account.mailboxes_diris read from/usr/local/lib/chatmaild/chatmail.ini(default/home/vmail/mail/<mail_domain>). - Active: chatmaild sets the mtime of the
passwordfile to the UTC day of the last login, so activity has one-day granularity. Nothing else is tracked. - New accounts: creation (birth) time of the account directory. Shows
n/aif the filesystem/kernel does not expose it. - IMAP:
doveadm who, so clients connecting through 443 (ALPN) are counted too, not only port 993. Idle connections hibernated by Dovecot (imap_hibernate_timeout) are counted as sessions but do not have animapprocess.
Install
git clone https://github.com/mrgluek/chatmail-relay-tools
cd chatmail-relay-tools
install -m 755 chatmail-stats.py /usr/local/bin/chatmail-stats
install -m 755 chatmail-ntfy-daily.sh /usr/local/bin/chatmail-ntfy-daily # optional
install -m 755 chatmail-kuma-push.sh /usr/local/bin/chatmail-kuma-push # optional
install -m 755 chatmail-apps-mirror.py /usr/local/bin/chatmail-apps-mirror # optional
install -m 755 transport-autofix.py /usr/local/bin/transport-autofix # optional
The ntfy and Kuma scripts call /usr/local/bin/chatmail-stats.
chatmail-stats
chatmail-stats # human-readable report
chatmail-stats --top 10 # + accounts with most IMAP sessions
chatmail-stats --json # for bots / monitoring
chatmail-stats --csv FILE # append one line (no addresses) — for cron
Example:
Accounts
total 1164
logged in today(UTC) 682
active 7d / 30d / 90d 1081 / 1164 / 1164
no login recorded 6
new 24h / 7d 261 / 888
IMAP (dovecot anvil)
connections 240
distinct accounts 187
distinct IPs 127
conns per account 1.28
imap processes 70 (570.8 MiB RSS)
RSS is summed per process and includes shared pages, so it overstates real
memory use; check free -m for the true picture.
Keep a history for trends:
echo '*/5 * * * * root /usr/local/bin/chatmail-stats --csv /var/log/chatmail-stats.csv' \
> /etc/cron.d/chatmail-stats
Daily summary to ntfy
cat > /etc/chatmail-ntfy.env <<'EOF'
NTFY_TOPIC=https://ntfy.example.org/chatmail-status
# NTFY_TOKEN=tk_... # if the topic is protected
# NTFY_TITLE=my.relay # default: mail_domain from chatmail.ini
# DISK_WARN=85 # disk % that raises the priority
EOF
chmod 600 /etc/chatmail-ntfy.env
chatmail-ntfy-daily
echo '55 23 * * * root /usr/local/bin/chatmail-ntfy-daily' > /etc/cron.d/chatmail-ntfy
Message:
Accounts: 1164 (+64)
Active today: 682
New 24h: 261
IMAP now: 240
Disk: 49%
The number in brackets is the net change since the previous run: new
registrations minus accounts removed by delete_inactive_users_after.
The first run has nothing to compare with and shows no change.
Uptime Kuma push monitors
Create one monitor of type Push per metric in Kuma and copy the token
(the part after /api/push/). Set the heartbeat interval a bit above the cron
interval (e.g. 600 s for a 5-minute cron).
cat > /etc/chatmail-kuma.env <<'EOF'
KUMA_URL=https://kuma.example.org
TOKEN_ACCOUNTS=...
TOKEN_IMAP=...
TOKEN_NEW24H=...
TOKEN_DISK=...
DISK_WARN=85
EOF
chmod 600 /etc/chatmail-kuma.env
echo '*/5 * * * * root /usr/local/bin/chatmail-kuma-push' > /etc/cron.d/chatmail-kuma
Each value is sent as ping, so Kuma draws it on the response-time chart
(ignore the "ms" label). Empty tokens are skipped. The disk monitor goes DOWN
when usage reaches DISK_WARN.
webxdc App Picker mirror
Delta Chat picks mini apps from two upstream sites: the
xdcget export (xdcget-lock.json, .xdc files,
icons) used by Desktop, and the store page opened
on Android. Where those are blocked, chatmail-apps-mirror keeps a copy of
both on the relay. Their file names do not overlap, so one URL serves every
client:
https://<relay>/apps/
Set it as the webxdc store URL in the client's experimental settings
(Desktop: Settings → Advanced → Experimental features), or just open it in a
browser. Desktop reads xdcget-lock.json from it; Android/iOS open it as the
store page.
What the copy changes:
main.jsloads the list and icons from/appsinstead of testrun.org.- In
xdcget-lock.jsonevery app'surlpoints at our copy of the.xdc. Desktop with a custom store URL downloads fromurl, which upstream is the developer's GitHub/Codeberg release. - Icons larger than 192 px are downscaled to 192 px in the same format (the
store shows them at 64 CSS px, so this covers 3x screens): about 2.4 MB of
icons become 1.2 MB. Needs
python3-pil; without it icons are served as is. Names and the lock are unchanged. - Desktop still loads icons from apps.testrun.org (hardcoded in the client); without access there the list shows no icons, but apps install.
How it is stored: each run builds a full release in
/var/lib/chatmail-apps-mirror/<time>/, then atomically switches the symlink
/var/www/html/apps to it. Unchanged files are hard-linked from the previous
release: .xdc files by name (the version is part of it, so they are not even
requested), icons and store files by content. apps.testrun.org re-dates all
its files on every export, so dates cannot tell what changed; comparing
content keeps our Last-Modified/ETag stable and lets browsers reuse cached
icons. A failed run leaves the previous copy in place;
the previous release is kept for rollback. About 120 MB.
The relay's nginx already serves /var/www/html and cmdeploy run copies its
web pages there without --delete, so no nginx change is needed and the
mirror survives redeploys.
apt install -y python3-pil # optional, for smaller icons
install -m 755 chatmail-apps-mirror.py /usr/local/bin/chatmail-apps-mirror
install -m 644 systemd/chatmail-apps-mirror.* /etc/systemd/system/
systemctl daemon-reload
systemctl start chatmail-apps-mirror # first sync, ~2 min
systemctl enable --now chatmail-apps-mirror.timer
journalctl -u chatmail-apps-mirror -n 5 # "167 apps, downloaded ..."
The public URL defaults to https://<mail_domain>/apps/ from chatmail.ini;
use --public-url / --dest for another place.
Privacy
Only counts are produced. Addresses appear only in the interactive --top
view on the server; they are never written to CSV, ntfy or Kuma.
License
MIT, see LICENSE.
transport-autofix
On relays where filtermail delivers outgoing mail (default_transport = lmtp-filtermail),
one bad destination stalls every group message that has a recipient there: filtermail runs
one worker per domain, postfix waits for all domains of a message, and you get
421 Transport busy, timed out while sending end of data after 600 s and duplicates.
A <domain> smtp: transport entry isolates such a domain: postfix smtp bounces 5xx
at once and defers only that recipient on TLS errors.
transport-autofix adds those entries automatically. Every 2 minutes it reads the new
postfix/lmtp-filtermail/lmtp journal lines, keeps 30 minutes of per-domain results and
routes a domain with at least 20 failed deliveries and at most 10 % successful ones for
24 hours. Healthy domains that only shared a timed-out group message keep a high success
rate and are left alone. Domains in /etc/postfix/transport, the discard list and the
relay's own domain are never touched. Every change goes to ntfy.
install -m 755 transport-autofix.py /usr/local/bin/transport-autofix
install -m 644 systemd/transport-autofix.service systemd/transport-autofix.timer /etc/systemd/system/
touch /etc/postfix/transport_auto && postmap /etc/postfix/transport_auto
postconf -e "transport_maps = hash:/etc/postfix/transport, hash:/etc/postfix/transport_discard_shared, hash:/etc/postfix/transport_auto"
postfix reload
printf 'MODE=dry-run\nNTFY_TOPIC=https://ntfy.example.org/chatmail-status\n' > /etc/transport-autofix.env
systemctl daemon-reload && systemctl enable --now transport-autofix.timer
transport-autofix --status
cmdeploy run rewrites main.cf, so add the transport_maps line to your post-deploy
steps. Start with MODE=dry-run (it only notifies what it would do), then switch to
MODE=apply. Settings and thresholds: see the header of transport-autofix.py.
transport-autofix --remove DOMAIN drops a route by hand.