ESXi虚拟环境下Tomcat向Postfix分发邮件速度过慢问题咨询

Hey there, let's dig into why your Tomcat-to-Postfix email queueing is running slow on your Fedora 27 VMs hosted on ESXi. I’ve worked through similar setup issues before, so here’s a structured approach to troubleshoot and fix this:

1. Start with Postfix's Core Configuration #

First, let’s rule out Postfix itself being the bottleneck.

• Pull up Postfix’s main config file with cat /etc/postfix/main.cf and check these critical settings:
  • smtpd_client_connection_count_limit: Ensure it’s not set too low (the default is 50, which should handle 20 emails without breaking a sweat)
  • local_recipient_maps: If this is misconfigured, Postfix might waste time on unnecessary lookups before accepting emails into the queue
  • smtpd_delay_reject: If this is disabled, Postfix could be rejecting connections prematurely, causing retries that slow things down
• Tail the Postfix logs in real-time while sending test emails: tail -f /var/log/maillog. Look for lines with connect from, status=queued, or explicit delay messages. If you see lags in handshake steps like EHLO or MAIL FROM, that points to a network or authentication issue.

2. Verify Network Latency Between VMs #

Even on the same ESXi host, network quirks can kill performance.

• Run a sustained ping test from your Tomcat VM to the Postfix VM: ping -c 100 <postfix-vm-ip>. If you see latency over 10ms or any packet loss, that’s a red flag for ESXi network congestion or misconfigured port groups.
• Test TCP connection speed with telnet <postfix-vm-ip> 25—if the connection takes more than a second to establish, check ESXi’s QoS rules and ensure both VMs are on the same, unthrottled port group.
• Use nc -zv <postfix-vm-ip> 25 for a quick connectivity check; any timeouts here confirm a network-level issue.

3. Audit Tomcat's Email Setup #

Tomcat’s mail implementation might be adding unnecessary overhead.

• Check your app’s mail session config (usually in context.xml or a dedicated properties file):
  • Enable connection pooling: Reopening an SMTP connection for every email is a huge slowdown. Look for properties like mail.smtp.connectionpool.size or use a library like Apache Commons Email that handles pooling automatically.
  • Tune timeout settings: Set mail.smtp.timeout and mail.smtp.connectiontimeout to reasonable values (e.g., 5000ms) to avoid unnecessary waiting.
  • Cache authentication: If you’re using mail.smtp.auth, make sure credentials aren’t being re-authenticated for every single email.
• Enable mail debug logging by adding mail.debug=true to your mail properties. Check Tomcat’s logs for delays in SMTP command execution—this will tell you exactly where the hold-up is happening.

4. Check VM Resource Constraints #

Even identical configs can hit resource limits on ESXi.

• Run htop on both VMs while sending test emails. Watch for:
  • High CPU usage from postfix or java processes (if either is maxed out, you need to allocate more vCPUs)
  • Spiking swap usage (insufficient RAM will slow down all operations, including queue writes)
• Test storage performance on the Postfix VM: The queue lives in /var/spool/postfix, so run dd if=/dev/zero of=/var/spool/postfix/testfile bs=1M count=100 to check write speeds. If it’s slower than 100MB/s, you might be dealing with ESXi thin provisioning lag or shared storage contention.
• Verify SELinux settings: Run getsebool -a | grep postfix to ensure postfix_can_sendmail is enabled, and check /var/log/audit/audit.log for any denials that could be blocking Postfix operations.

5. Isolate the Issue with a Minimal Test #

Rule out your Tomcat app code as the culprit.

• Write a simple bash script to send 20 emails directly to Postfix, bypassing Tomcat:

  for i in {1..20}; do
      echo -e "Subject: Test Email $i\n\nThis is test email $i" | sendmail your-email@example.com
  done
  

• If this script runs quickly (under 1 second), the problem is definitely in your Tomcat app’s mail implementation. If it’s still slow, focus back on Postfix or network settings.

Once you work through these steps, you’ll pinpoint exactly where the bottleneck is. For example, enabling connection pooling in Tomcat can cut queueing time to near-instant for bulk emails, while fixing ESXi network congestion will resolve latency-related delays.

内容的提问来源于stack exchange，提问作者Michael