HistoryIndexCacheSize=32M

PID in your logs. If you see it restarting frequently, it may be crashing due to a specific malformed item or excessive load Increase the for the specific service to get more detail: zabbix_server -R log_level_increase="preprocessing manager" 4. Verification & Clean Restart

How to diagnose (ordered, practical)

Are you running a or a Zabbix Proxy when this happens?

A: The complexity of HA environments can expose timing-sensitive issues that might not appear in single-server deployments. However, the underlying causes—resource limits and workload spikes—remain the same. Ensure HA nodes have identical and adequate resource configurations.

# On Zabbix server grep -i "broken pipe" /var/log/zabbix/zabbix_server.log

If the reading process (consumer) cannot consume data fast enough to clear the buffer, the operating system may close the socket to prevent a deadlock, resulting in a broken pipe.

: Ensure these are high enough to support your Zabbix CacheSize settings.

Look for "Out of memory," "Connection refused," or "Slow query" messages immediately preceding the IPC error. 3. Review Database Performance

Increase values gradually (e.g., double each). Restart Zabbix server: systemctl restart zabbix-server .

If iowait is >10% or your disks are saturated, Zabbix processes might be blocking on disk writes.

sudo systemctl daemon-reload sudo systemctl restart zabbix-server Use code with caution. 2. Expand the Zabbix History Cache Size

This forces Zabbix to recreate all shared memory segments and message queues from scratch.

zabbix soft nofile 65536 zabbix hard nofile 65536 zabbix soft nproc 65536 zabbix hard nproc 65536

When a "broken pipe" (SIGPIPE) occurs, it means the reader process died, crashed, or timed out while the writer process was still transmitting data. Root Causes and Solutions 1. Preprocessing Worker Crashes (Most Common)

The history syncers write to the database. A slow DB will cause broken pipes.