Enabling Dovecot Full‑Text Search with Solr: Setup, Querying, and Trade‑offs

Problem: Slow IMAP SEARCH on large mailboxes

When a Dovecot server must scan every message to satisfy a TEXT or HEADER SEARCH, CPU usage spikes and latency grows with mailbox size. Users notice delayed results, especially in shared or archive folders.

Thesis: Off‑load indexing to an external Solr instance via the fts_solr plugin keeps search responsive while moving the indexing workload away from the mail server.

Configuration Steps

1. Install the Solr plugin package (usually dovecot-fts-solr) and ensure a Solr core named dovecot is reachable at http://solr-host:8983/solr/dovecot.
2. Edit /etc/dovecot/conf.d/10-mail.conf (or a local override) and add the plugin to the mail_plugins line:

   mail_plugins = $mail_plugins fts fts_solr

3. Create a file /etc/dovecot/conf.d/20-fts_solr.conf with:

   plugin {
     fts = solr
     fts_solr_url = http://solr-host:8983/solr/dovecot
     fts_solr_timeout = 5
     # Index Subject, From, To, Body
     fts_solr_fields = subject from to body
   }

4. Restart Dovecot: sudo systemctl restart dovecot (requires root or sudo).

Indexing Workflow and Query Example

After restart, Dovecot indexes new messages as they are delivered. For a manual test you can inject a message with doveadm:

# Replace USER with a real account
sudo -u dovecot doveadm mailbox create -u USER Test
sudo -u dovecot doveadm save -u USER Test 

Then, from any IMAP client (or via telnet/openssl) log in and run:

a001 LOGIN USER password
a002 SEARCH TEXT \"Alpha\"

If indexing succeeded, the server returns a list of matching UIDs, e.g., * SEARCH 12345. The same query can be checked directly in Solr:

curl -s "http://solr-host:8983/solr/dovecot/select?q=Alpha&wt=json" | jq '.response.numFound'

Trade‑offs and Limitations

• Operational dependency: If the Solr node is unreachable, Dovecot logs a warning and returns empty results for SEARCH; there is no automatic fallback to local scanning.
• Schema alignment: The fts_solr plugin expects fields named exactly as configured (subject, from, to, body) with text_general type. Changing the Solr schema requires matching those names and restarting Dovecot.
• Latency: Search latency now includes the network round‑trip to Solr plus Solr’s query time. For a local LAN this is usually <10 ms, but a remote Solr cluster can add noticeable delay.

Verification and Next Steps

1. Check Dovecot logs after restart for lines like fts_solr: connected to Solr at http://solr-host:8983/solr/dovecot.
2. Use telnet or openssl to connect to the IMAP port, log in, and issue SEARCH TEXT \"test\"; a non‑empty UID list confirms end‑to‑end flow.
3. Query Solr directly (as shown above) to verify that the document sent by Dovecot appears in the index.
4. Monitor CPU usage on the Dovecot host before and after enabling fts_solr (e.g., with top or pidstat) to observe the expected reduction.

Once verified, you can adjust fts_solr_timeout or tune Solr’s heap and commit settings to balance freshness versus load. Remember to include the Solr host in your monitoring and alerting so search failures are caught early.