Emergency procedures: Difference between revisions

From VoIPmonitor.org
(Add dmesg -T as critical first diagnostic step for MEMORY IS FULL errors, add NFS network latency diagnostics)
(Review: použití šablon {{Note}} a {{Warning}} místo prostého textu)
(5 intermediate revisions by the same user not shown)
Line 1: Line 1:
{{DISPLAYTITLE:Emergency Procedures & System Recovery}}
= Diagnosing Database Bottlenecks Using Sensor RRD Charts =


'''This guide covers emergency procedures for recovering your VoIPmonitor system from critical failures, including runaway processes, high CPU usage, and system unresponsiveness.'''
If your VoIPmonitor GUI becomes unresponsive or PHP processes are being terminated by the OOM (Out of Memory) killer, the root cause may be a '''database performance bottleneck''', not a PHP configuration issue.


== Emergency: VoIPmonitor Process Consuming Excessive CPU or System Unresponsive ==
This guide explains how to use the sensor's RRD (Round-Robin Database) charts to identify whether the database server is the limiting factor.


When a VoIPmonitor process consumes excessive CPU (e.g., ~3000% or more) or causes the entire system to become unresponsive, follow these immediate steps:
== Symptoms of Database Bottlenecks affecting the GUI ==


=== Immediate Action: Force-Terminate Runaway Process ===
* GUI becomes extremely slow or unresponsive during peak hours
* PHP processes are killed by the OOM killer on the GUI server
* Dashboard and CDR views take a long time to load
* Alerts and reports fail during high traffic periods
* System appears fine during off-peak hours but degrades during peak usage


If the system is still minimally responsive via SSH or requires out-of-band management (iDRAC, IPMI, console):
{{Note|These symptoms often occur when the GUI server is waiting for database queries to complete, causing PHP processes to pile up and consume excessive memory.}}


;1. Identify the Process ID (PID):
== Understanding Sensor RRD Charts ==
<syntaxhighlight lang="bash">
# Using htop (if available)
htop


# Or using ps
The VoIPmonitor sensor generates performance charts (RRD files) that track system metrics over time. These charts are accessible through the GUI and provide visual indicators of where bottlenecks are occurring.
ps aux | grep voipmonitor
</syntaxhighlight>


Look for the voipmonitor process consuming the most CPU resources. Note down the PID (process ID number).
To access sensor RRD charts:
# Navigate to '''Settings > Sensors''' in the GUI
# Click the graph icon next to the sensor
# Select the time range covering the problematic peak hours


;2. Forcefully terminate the process:
== Diagnostic Flowchart ==
<syntaxhighlight lang="bash">
kill -9 <PID>
</syntaxhighlight>


Replace <PID> with the actual process ID number identified in step 1.
<kroki lang="mermaid">
flowchart TD
    A[GUI Unresponsive / OOM Errors] --> B{Check Sensor RRD Charts}
    B --> C{SQL Cache Growing<br/>During Peak Hours?}
    C -->|No| D[Issue is NOT<br/>database bottleneck]
    C -->|Yes| E[Database Bottleneck<br/>Confirmed]
    E --> F{Identify<br/>Bottleneck Type}
    F --> G{mysqld CPU<br/>near 100%?}
    G -->|Yes| H[CPU Bottleneck]
    H --> I[Add CPU cores<br/>or upgrade CPU]
    G -->|No| J{Buffer pool full?<br/>Swap usage?}
    J -->|Yes| K[Memory Bottleneck]
    K --> L[Add RAM<br/>Tune innodb_buffer_pool_size]
    J -->|No| M{High iowait?<br/>Magnetic disks?}
    M -->|Yes| N[Storage I/O Bottleneck]
    N --> O[Upgrade to SSD/NVMe]


;3. Verify system recovery:
    style A fill:#f9f,stroke:#333
<syntaxhighlight lang="bash">
    style E fill:#ff9,stroke:#333
# Check CPU usage has returned to normal
    style H fill:#f96,stroke:#333
top
    style K fill:#f96,stroke:#333
    style N fill:#f96,stroke:#333
</kroki>


# Check if the process was terminated
== Diagnostic Step 1: Look for Growing SQL Cache ==
ps aux | grep voipmonitor
</syntaxhighlight>


The system should become responsive again immediately after the process is killed. CPU utilization should drop significantly.
The most critical indicator of a database bottleneck is '''growing SQL cache''' or '''SQL cache files''' during peak hours.


=== Optional: Stop and Restart the Service (for persistent issues) ===
{| class="wikitable"
 
|-
If the problem persists or the service needs to be cleanly restarted:
! Metric !! What to Look For !! Indicates
|-
| '''SQL Cache''' || Consistently increasing during peak hours, never decreasing || Database cannot keep up with insert rate
|-
| '''SQL Cache Files''' || Growing over time during peak usage || Database buffer pool too small or storage too slow
|-
| '''CPU Load (mysqld)''' || Near 100% during peak hours || CPU bottleneck on database server
|-
| '''Disk I/O (mysql)''' || High or saturated during peak hours || Storage bottleneck (magnetic disks instead of SSDs)
|}


<syntaxhighlight lang="bash">
If you see SQL cache or SQL cache files growing consistently during peak traffic periods, the database server is the bottleneck.
# Stop the voipmonitor service
systemctl stop voipmonitor


# Verify no zombie processes remaining
== Diagnostic Step 2: Determine the Bottleneck Type ==
killall voipmonitor


# Restart the service
After identifying that the database is the issue, determine which resource is the limiting factor:
systemctl start voipmonitor


# Verify service status
=== CPU Bottleneck ===
systemctl status voipmonitor
* Check database CPU usage during peak hours
</syntaxhighlight>
* If mysqld is at or near 100% CPU, you need more CPU cores or faster CPUs


'''Caution:''' When using <code>systemd</code> service management, avoid using the deprecated <code>service</code> command as it can cause systemd to lose track of the daemon. Always use <code>systemctl</code> commands or direct process commands like <code>killall</code>.
=== Memory Bottleneck ===
* Check if SQL cache grows because buffer pool is too small
* Database runs out of RAM for caching, forcing disk reads
* The SQL cache chart shows a pattern of filling up and staying full


=== Root Cause Analysis: Why Did the CPU Spike? ===
=== Storage I/O Bottleneck (Most Common) ===
* High disk I/O wait times for mysqld process
* Disk latency (iowait) increases during peak hours
* Database storage on magnetic disks (e.g., 10K SAS) instead of SSD/NVMe
* SQL cache grows because data cannot be written/read fast enough


After recovering the system, investigate the root cause to prevent recurrence. Common causes include:
== Solutions for Database Performance Bottlenecks ==


;SIP REGISTER Flood / Spaming Attack
=== Solution 1: Add More RAM to the Database Server ===
Massive volumes of SIP REGISTER messages from malicious IPs can overwhelm the VoIPmonitor process.


* '''Detection:''' Check recent alert triggers in the VoIPmonitor GUI > Alerts > Sent Alerts for SIP REGISTER flood alerts
This is often the most effective fix for memory-related bottlenecks.
* '''Immediate mitigation:''' Block attacker IPs at the network edge (SBC, firewall, iptables)
* '''Long-term prevention:''' Configure anti-fraud rules with custom scripts to auto-block, see [[Anti-fraud#SIP REGISTER Flood/Attack|SIP REGISTER Flood Mitigation]]


;Packet Capture Overload (pcapcommand)
{| class="wikitable"
The <code>pcapcommand</code> feature forks a program for ''every'' call, which can generate up to 500,000 interrupts per second.
|-
! Current RAM !! Recommended Upgrade !! Expected Impact
|-
| 32GB || 64GB or 128GB || Significantly reduces cache growth
|-
| 64GB || 128GB or 256GB || Handles much higher peak loads
|-
| 128GB || 256GB || Suitable for large deployments
|}


* '''Detection:''' Check <code>/etc/voipmonitor.conf</code> for a <code>pcapcommand</code> line
After adding RAM, tune <code>innodb_buffer_pool_size</code> in your MySQL configuration:
* '''Immediate fix:''' Comment out or remove the <code>pcapcommand</code> directive and restart the service
* '''Alternative:''' Use the built-in cleaning spool functionality (<code>maxpoolsize</code>, <code>cleanspool</code>) instead
 
;Excessive RTP Processing Threads
High concurrent call volumes can overload RTP processing threads.
 
* '''Detection:''' Check performance logs for high <code>tRTP_CPU</code> values (sum of all RTP threads)
* '''Mitigation:'''
  <pre>callslimit = 2000  # Limit max concurrent calls</pre>
 
;Audio Feature Overhead
Silence detection and audio conversion are CPU-intensive operations.
 
* '''Detection:''' Check if <code>silencedetect</code> or <code>saveaudio</code> are enabled
* '''Mitigation:'''
  <pre>
  silencedetect = no
  # saveaudio = wav  # Comment out if not needed
  </pre>
 
See [[Scaling|Scaling and Performance Tuning]] for detailed performance optimization strategies.
 
=== Preventive Measures ===
 
Once the root cause is identified, implement these preventive configurations:
 
;Monitor CPU Trends:
Use [[Collectd_installation|collectd]] or your existing monitoring system to track CPU usage over time and receive alerts before critical thresholds are reached.
 
;Anti-Fraud Auto-Blocking:
Configure [[Anti-fraud|Anti-Fraud rules]] with custom scripts to automatically block attacker IPs when a flood is detected. See the [[Anti-fraud|Anti-Fraud documentation]] for PHP script examples using iptables or ipset.
 
;Network Edge Protection:
Block SIP REGISTER spam and floods at your network edge (SBC, firewall) before traffic reaches VoIPmonitor. This provides better performance and reduces CPU load on the monitoring system.
 
== Emergency: GUI and CLI Frequently Inaccessible Due to Memory Exhaustion ==
 
When the VoIPmonitor GUI and CLI become frequently inaccessible or the server becomes unresponsive due to Out of Memory (OOM) conditions, follow these steps to identify and resolve the issue.
 
=== Diagnose OOM Events ===
 
The Linux kernel out-of-memory (OOM) killer terminates processes when RAM is exhausted.
 
;Check the kernel ring buffer for OOM events:
<syntaxhighlight lang="bash">
dmesg -T | grep -i killed
</syntaxhighlight>
 
If you see messages like "Out of memory: Kill process" or "invoke-oom-killer", your system is running out of physical RAM.
 
=== Immediate Relief: Reduce Memory Allocation ===
 
Reduce memory consumption by tuning both MySQL and VoIPmonitor parameters.
 
;1. Reduce MySQL Buffer Pool Size:
 
Edit the MySQL configuration file (typically <code>/etc/my.cnf.d/mysql-server.cnf</code> or <code>/etc/mysql/my.cnf</code> for Debian/Ubuntu):


<syntaxhighlight lang="ini">
<syntaxhighlight lang="ini">
[mysqld]
# /etc/mysql/my.cnf
# Reduce from 8GB to 6GB (adjust based on available RAM)
# Set to 50-70% of total RAM on dedicated database server
innodb_buffer_pool_size = 6G
innodb_buffer_pool_size = 128G
</syntaxhighlight>
</syntaxhighlight>


A good starting point is <code>innodb_buffer_pool_size = RAM * 0.5 - max_buffer_mem * 0.8</code>. For example, on a 16GB server with 8GB allocated to max_buffer_mem, set innodb_buffer_pool_size to approximately 6GB.
For more tuning guidance, see [[Scaling#Optimizing_Database_Performance_.28MySQL.2FMariaDB.29|Scaling - Database Performance]].


;2. Reduce VoIPmonitor Buffer Memory:
{{Warning|1=Do NOT reduce <code>innodb_buffer_pool_size</code> on the GUI server when the database is the bottleneck. This will make the problem worse.}}


Edit <code>/etc/voipmonitor.conf</code> and decrease the <code>max_buffer_mem</code> value:
=== Solution 2: Upgrade Database Storage to SSD/NVMe ===


<syntaxhighlight lang="ini">
If your database storage is on magnetic disks (e.g., 10K SATA or SAS), upgrading to SSDs is often the single most effective improvement.
[general]
# Reduce from 8000 to 6000 (adjust based on available RAM)
max_buffer_mem = 6000
</syntaxhighlight>


The <code>max_buffer_mem</code> parameter limits the maximum RAM allocation for the packet buffer. Typical values range from 2000-8000 MB depending on traffic volume and call rates.
{| class="wikitable"
 
|-
;3. Restart the affected services:
! Current Storage !! Recommended Upgrade !! Expected Speedup
 
<syntaxhighlight lang="bash">
systemctl restart mysqld
systemctl restart voipmonitor
</syntaxhighlight>
 
Monitor the system to confirm stability.
 
=== Long-term Solution: Increase RAM ===
 
For sustained production operation, increase the server's physical RAM:
 
* '''Minimum''': Add at least 16 GB of additional RAM to eliminate OOM conditions
* '''Performance benefit''': After the RAM upgrade, you can safely increase <code>innodb_buffer_pool_size</code> to improve MySQL performance
* '''Recommended settings''': Set <code>innodb_buffer_pool_size</code> to 50-70% of total RAM and <code>max_buffer_mem</code> based on your traffic requirements
 
See [[Sniffer_configuration#max_buffer_mem|Sniffer Configuration]] for details on VoIPmonitor memory settings.
 
== Emergency: Diagnosing System Hangs and Collecting Core Dump Evidence ==
 
When the VoIPmonitor system hangs, packet buffer (heap) spikes to 100%, and a single CPU core is pegged at 100%, you need to diagnose the issue and collect evidence for developer analysis before restarting.
 
### Identify the Problematic Thread
 
Use the Manager API to identify which sniffer thread is consuming excessive CPU resources.
 
<syntaxhighlight lang="bash">
# Query thread statistics from the sensor
echo 'sniffer_threads' | nc <sensor_ip> 5029
</syntaxhighlight>
 
Replace <sensor_ip> with the actual IP address of your VoIPmonitor sensor. Look for a thread showing approximately 100% CPU usage. This indicates the specific processing thread that is causing the hang.
 
### Generate Core Dump for Developer Analysis
 
If a thread is pegged at 100% and the system needs to be analyzed by VoIPmonitor developers, generate a core dump before restarting:
 
;1. Find the VoIPmonitor process ID (PID):
<syntaxhighlight lang="bash">
ps aux | grep voipmonitor | grep -v grep
</syntaxhighlight>
 
;2. Attach to the process with gdb and generate a core dump:
<syntaxhighlight lang="bash">
gdb -p <PID_of_voipmonitor>
# Within gdb, generate the core dump
gcore <output_file>
 
# Example:
gdb -p 12345
(gdb) gcore /tmp/voipmonitor_hang.core
</syntaxhighlight>
 
The core dump file provides developers with a complete snapshot of the process state at the moment of the hang, including memory, registers, and stack traces.
 
;3. Detach from gdb and quit:
<syntaxhighlight lang="bash"> detach
quit
</syntaxhighlight>
 
### Restore Service and Collect Evidence
 
After collecting the diagnostic evidence, restart the service to restore operation:
 
<syntaxhighlight lang="bash">
systemctl restart voipmonitor
</syntaxhighlight>
 
Provide the following files to VoIPmonitor support for analysis:
 
* Core dump file (from gcore command)
* Thread statistics output (from sniffer_threads command)
* Performance logs (/var/log/syslog showing the hang period)
* Configuration file (/etc/voipmonitor.conf)
 
'''Important:''' Core dump files can be very large (several GB depending on max_buffer_mem). Ensure you have sufficient disk space and consider compressing the file before transferring it to support.
 
== Emergency: System Freezes on Every Update Attempt ==
 
If the VoIPmonitor sensor becomes unresponsive or hangs each time you attempt to update it through the Web GUI:
 
;1. SSH into the sensor host
;2. Execute the following commands to forcefully stop and restart:
<syntaxhighlight lang="bash">
killall voipmonitor
systemctl stop voipmonitor
systemctl start voipmonitor
</syntaxhighlight>
 
This sequence ensures zombie processes are terminated, systemd is fully stopped, and a clean service restart occurs. Verify the sensor status in the GUI to confirm it is responding correctly.
 
== Emergency: Binary Not Found After Crash ==
 
If the VoIPmonitor service fails to start after a crash with error "Binary not found" for <code>/usr/local/sbin/voipmonitor</code>:
 
;1. Check for a renamed binary:
<syntaxhighlight lang="bash">
ls -l /usr/local/sbin/voipmonitor_*
</syntaxhighlight>
 
The crash recovery process may have renamed the binary with an underscore suffix.
 
;2. If found, rename it back:
<syntaxhighlight lang="bash">
mv /usr/local/sbin/voipmonitor_ /usr/local/sbin/voipmonitor
</syntaxhighlight>
 
;3. Restart the service:
<syntaxhighlight lang="bash">
systemctl start voipmonitor
systemctl status voipmonitor
</syntaxhighlight>
 
Verify the service starts correctly.
 
== Out-of-Band Management Scenarios ==
 
When the system is completely unresponsive and cannot be accessed via SSH:
 
* '''Use your server's out-of-band management system:'''
  * Dell iDRAC
  * HP iLO
  * Supermicro IPMI
  * Other vendor-specific BMC/management tools
 
* '''Actions available via OBM:'''
  * Access virtual console (KVM-over-IP)
  * Send NMI (Non-Maskable Interrupt) for system dump
  * Force power cycle
  * Monitor hardware health
 
See [[Sniffer_troubleshooting|Sniffer Troubleshooting]] for more diagnostic procedures.
 
== Emergency: Service Restart Loop with "packetbuffer: MEMORY IS FULL" and "Cannot bind to port" ==
 
If the VoIPmonitor service enters a restart loop, logging <code>packetbuffer: MEMORY IS FULL</code> and displaying <code>Cannot bind to port [5029]</code> errors, the issue can have '''multiple root causes'''. The "MEMORY IS FULL" error message is ambiguous and can indicate either RAM exhaustion or disk I/O bottleneck.
 
=== Critical: Distinguish Between RAM and Disk I/O Issues ===
 
The symptoms appear identical, but the root causes and solutions are different:
 
{|
|-
|-
! style="background:#ffc107;" | RAM-Based Memory Issue
| 10K RPM SATA HDD || NVMe SSD array || 10-50x faster I/O
! style="background:#ffc107;" | Disk I/O Performance Issue
|-
|-
| Memory buffer fills due to excessive concurrent calls or traffic floods
| 10K RPM SAS HDD || Enterprise SSD (SAS/SATA) || 5-20x faster I/O
| Memory buffer fills because disk cannot write fast enough to drain it
|-
|-
| Solution: Increase <code>max_buffer_mem</code>, enable <code>packetbuffer_compress</code>, or limit concurrent calls
| Older SSD || Modern NVMe (PCIe 4.0+) || 2-5x faster I/O
| Solution: Upgrade storage, move spool to faster disk, or resolve I/O bottleneck
|}
|}


=== Step 0: Check Kernel Messages for Storage Errors (Critical First Step!) ===
For high-traffic deployments, '''NVMe storage is recommended for the database host'''.


Before investigating performance issues, check the kernel message buffer for storage hardware or filesystem errors. This is the '''first diagnostic step''' to distinguish between hardware/failure problems and performance bottlenecks.
See [[Hardware#Database_Storage|Hardware - Storage Selection]] for detailed recommendations.


;1. Check kernel messages for storage errors:
=== Solution 3: Temporary Mitigation - Schedule Alerts/Reports Outside Peak Hours ===
<syntaxhighlight lang="bash">
# Check the kernel message buffer for storage-related errors
dmesg -T | grep -i -E "i/o error|disk|storage|filesystem|ext4|xfs|nfs|scsi"
</syntaxhighlight>
 
* '''What to look for:'''
  * I/O errors (e.g., "Buffer I/O error", "critical medium error")
  * Filesystem errors (e.g., "EXT4-fs error", "XFS error")
  * NFS-specific errors (e.g., "NFS: server not responding", "NFS: device not ready")
  * SCSI/SATA errors (e.g., "Task abort", "Device failed")
  * ATA SMART errors indicating disk degradation
 
;2. If kernel errors are present:
** This indicates a hardware or filesystem issue, not a performance bottleneck
** Solutions depend on the specific error:
  * Replace failing disk hardware
  * Repair filesystem (fsck)
  * Resolve NFS connectivity issues (network, server availability)
  * Check RAID controller for failures
  * Fix underlying kernel/storage configuration issues
 
;3. If kernel messages are clean (no errors):
** Proceed to '''Step 1''' below to investigate disk I/O performance bottlenecks
 
For more detailed kernel event investigation, use:
<syntaxhighlight lang="bash">
# View all recent kernel messages with timestamps
dmesg -T | tail -100


# Filter for time range (example: last 1 hour)
If you cannot immediately upgrade the database server hardware, temporarily reduce the load by scheduling intensive tasks during off-peak hours.
journalctl -k --since "1 hour ago"
</syntaxhighlight>


=== Step 1: Check for Disk I/O Bottleneck (Performance Issue) ===
'''1. Disable or reduce alert frequency''' during peak hours:
* Navigate to '''GUI > Alerts'''
* Temporarily disable high-frequency alerts
* Set alerts to run during off-peak periods (e.g., 2am-4am)


If <code>dmesg -T</code> shows no storage errors (Step 0), the issue is likely a performance bottleneck in the storage subsystem. Check for disk I/O problems on the spool directory (typically <code>/var/spool/voipmonitor</code>).
'''2. Schedule reports outside peak usage:'''
* Navigate to '''GUI > Reports'''
* Configure scheduled reports for off-peak hours
* Avoid generating reports during the busiest part of the day


;1. Monitor disk utilization with iostat:
'''3. Reduce dashboard complexity''' during peak hours:
<syntaxhighlight lang="bash">
* Simplify dashboards that query large ranges of data
# Monitor disk I/O in real-time (1-second intervals)
* Avoid "All time" statistics during peak loads
iostat -x 1
* Use cached dashboards or static displays when possible
</syntaxhighlight>
* '''What to look for:''' A value near 100% in the <code>%util</code> column indicates the disk is operating at maximum capacity
* '''Symptoms:** High %util, high await (average wait time), or high queue depth


;2. Perform a write speed test to the spool directory:
=== Solution 4: Consider Component Separation ===
<syntaxhighlight lang="bash">
# Test sequential write speed (adjust count based on available disk space)
# Note: dd test uses O_DIRECT to bypass cache for accurate measurement
dd if=/dev/zero of=/var/spool/voipmonitor/testfile bs=1M count=1024 oflag=direct conv=fdatasync


# Clean up test file
If the database server is a bottleneck and upgrading is not feasible, consider moving to a dedicated database architecture.
rm /var/spool/voipmonitor/testfile
</syntaxhighlight>
* '''Interpretation:''' A very slow write speed (e.g., less than 50 MB/s on HDDs or significantly lower than expected SSD speed) confirms a storage bottleneck
* For SSD/NVMe, expect 400+ MB/s sequential writes
* For HDDs, expect 80-150 MB/s sequential writes (7200 RPM)


;3. Check for I/O wait (Linux monitoring):
In a component separation deployment (see [[Scaling#Scaling_Through_Component_Separation|Scaling - Component Separation]]):
<syntaxhighlight lang="bash">
* '''Host 1:''' Dedicated database server with maximum RAM and SSD/NVMe storage
# Check if the system is spending significant time waiting for I/O
* '''Host 2:''' GUI web server
# High 'wa' (wait) percentage indicates disk bottleneck
* '''Host 3:''' Sensor(s)
top
# or
vmstat 1
</syntaxhighlight>
* Look for high <code>%wa</code> (I/O wait) in the CPU section


=== Step 2: Resolve Disk I/O Bottleneck ===
This allows you to independently scale the database with more powerful hardware without affecting the GUI.


If disk I/O tests confirm the issue:
== Common Pitfalls to Avoid ==
 
* '''Option 1: Upgrade storage hardware'''
  ** Move <code>/var/spool/voipmonitor</code> to a faster local SSD or NVMe drive
  ** Consider RAID 10 for better performance and redundancy
  ** If using NFS, move spool to local storage instead of network-mounted filesystem
 
* '''Option 2: Tune storage configuration'''
  ** Check if the disk is operating in degraded mode (RAID rebuild in progress)
  ** Verify the storage controller firmware is up to date
  ** Disable unnecessary monitoring or indexing (e.g., updatedb, antivirus scanning) on the spool directory
 
* '''Option 2a: NFS Network Storage Performance'''
  If <code>/var/spool/voipmonitor</code> is mounted on NFS:
  ** Check network latency to NFS server:
    <syntaxhighlight lang="bash">
    # Ping test to NFS server
    ping -c 10 <nfs_server_ip>
 
    # Measure NFS-specific latency/mount stats
    # Requires nfsiostat from nfs-utils package
    nfsiostat 1
    </syntaxhighlight>
  ** Check NFS server response time and network congestion
  ** Consider upgrading network (e.g., 10GbE) for higher NFS throughput
  ** Use TCP mount options for reliability (e.g., <code>mount -t nfs -o tcp</code>)
  ** Verify NFS server has sufficient disk I/O performance
  ** If NFS is the bottleneck, move spool directory to local SSD storage
 
* '''Option 3: Move spool directory to faster volume'''
  <syntaxhighlight lang="bash">
  # Stop service
  systemctl stop voipmonitor
 
  # Mount faster disk to /var/spool/voipmonitor
  # Or create symlink:
  mv /var/spool/voipmonitor /var/spool/voipmonitor.backup
  ln -s /path/to/fast/disk/voipmonitor /var/spool/voipmonitor
 
  # Restart service
  systemctl start voipmonitor
  </syntaxhighlight>
 
For detailed disk performance benchmarking, see [[IO_Measurement|I/O Performance Measurement]] for advanced testing with <code>fio</code> and <code>ioping</code>.
 
=== Step 3: Check for RAM-Based Memory Issue ===
 
If disk I/O is healthy but the error persists, the issue is RAM-based memory exhaustion.
 
;1. Check RAM allocation:
<syntaxhighlight lang="bash">
# Check current memory usage
free -h
</syntaxhighlight>
 
;2. Increase memory buffer limits:
Edit <code>/etc/voipmonitor.conf</code>:


{| class="wikitable" style="background:#fff3cd; border:1px solid #ffc107;"
{| class="wikitable" style="background:#fff3cd; border:1px solid #ffc107;"
|-
|-
! colspan="2" style="background:#ffc107;" | Recommended Values for "MEMORY IS FULL" Errors
! colspan="2" style="background:#ffc107;" | Incorrect Solutions When Database is the Bottleneck
|-
|-
| '''ringbuffer''' || For very high traffic (>200Mbps) or severe packet loss scenarios, increase to 2000 MB (maximum allowed). Default is 50 MB, recommended for >100Mbit traffic is 500 MB.
| style="vertical-align: top;" | '''Reducing PHP memory_limit'''
| This does NOT fix the root cause. PHP waits for the database; less memory means processes crash sooner.
|-
|-
| '''max_buffer_mem''' || For high concurrent call loads (5000+ calls) or persistent buffer issues, increase to 8000 MB. Default is 2000 MB, typical tuning is 4000 MB for moderate loads.
| style="vertical-align: top;" | '''Tuning PHP-FPM worker counts'''
| More workers will pile up waiting for slow database queries, consuming even more memory.
|-
|-
| '''packetbuffer_compress''' || Enable if RAM is constrained (increases CPU usage to reduce memory footprint).
| style="vertical-align: top;" | '''Reducing innodb_buffer_pool_size'''
| This makes the database slower, not faster. It causes more disk I/O and longer query times.
|-
| style="vertical-align: top;" | '''Adding RAM to the GUI server'''
| If the bottleneck is the database, adding RAM to the GUI won't help. The database is the limiting factor.
|}
|}


<syntaxhighlight lang="ini">
== Verification Checklist ==
[general]
# HIGH TRAFFIC CONFIGURATION - Prevent "MEMORY IS FULL" errors
# Max ringbuffer for very high traffic traffic/serious packet loss
ringbuffer = 2000
 
# Increase buffer memory for high concurrent call loads
max_buffer_mem = 8000
 
# Enable compression to save RAM at CPU cost
packetbuffer_compress = yes
 
# Optional: Limit concurrent calls to prevent overload
callslimit = 2000
</syntaxhighlight>
 
'''Alternative: Moderate Traffic Configuration'''
<syntaxhighlight lang="ini">
[general]
# For moderate traffic (100-200 Mbit, 2000-5000 concurrent calls)
ringbuffer = 500
max_buffer_mem = 4000
packetbuffer_compress = yes
</syntaxhighlight>
 
;3. Restart and monitor:
<syntaxhighlight lang="bash">
systemctl restart voipmonitor
journalctl -u voipmonitor -f
</syntaxhighlight>
 
See [[Sniffer_configuration#max_buffer_mem|Sniffer Configuration]] for more memory tuning options.
 
=== Step 4: Alternative Root Cause - Adaptive Jitterbuffer Overload ===
 
If the "packetbuffer: MEMORY IS FULL" and "HEAP FULL" errors occur even after adjusting <code>max_buffer_mem</code>, the issue may be caused by the adaptive jitterbuffer feature consuming excessive memory during processing. The adaptive jitterbuffer (which simulates jitter up to 500ms) is CPU and memory-intensive and can trigger heap exhaustion on high-traffic systems.
 
;1. Check if jitterbuffer_adapt is enabled:
<syntaxhighlight lang="bash">
# Check voipmonitor.conf for jitter buffer settings
grep jitterbuffer /etc/voipmonitor.conf
</syntaxhighlight>
 
If <code>jitterbuffer_adapt = yes</code> is set, this features may be causing the memory exhaustion.
 
;2. Disable adaptive jitterbuffer:
Edit <code>/etc/voipmonitor.conf</code> and set:
<syntaxhighlight lang="ini">
[general]
# Disable adaptive jitterbuffer to prevent memory/CPU exhaustion
jitterbuffer_adapt = no
</syntaxhighlight>
 
;3. Restart the service:
<syntaxhighlight lang="bash">
systemctl restart voipmonitor
</syntaxhighlight>
 
;4. Verify the error is resolved:
<syntaxhighlight lang="bash">
# Monitor for MEMORY IS FULL errors
journalctl -u voipmonitor -f
</syntaxhighlight>
 
'''Important Trade-offs:'''
 
* Disabling <code>jitterbuffer_adapt</code> removes the CPU/memory overhead but also disables <code>MOS_adaptive</code> score calculation
* Fixed jitterbuffer modes (<code>jitterbuffer_f1</code> for 50ms, <code>jitterbuffer_f2</code> for 200ms) remain available and consume significantly less resources
* If MOS quality scoring is required, consider using <code>jitterbuffer_f2 = yes</code> instead
 
This solution is particularly effective when the system crashes with both "MEMORY IS FULL" and "HEAP FULL" errors simultaneously, indicating the adaptive jitterbuffer heap is overflowing during real-time packet processing.
 
=== Step 5: Clear Stale Port 5029 Bindings ===
 
The "Cannot bind to port [5029]" error occurs when a zombie process still holds the Manager API port. This prevents clean restarts.
 
<syntaxhighlight lang="bash">
# Force kill all VoIPmonitor processes
killall -9 voipmonitor
 
# Ensure service is stopped
systemctl stop voipmonitor
 
# Verify no processes are running
ps aux | grep voipmonitor
 
# Restart service
systemctl start voipmonitor
</syntaxhighlight>
 
After clearing zombie processes and addressing the root cause (I/O or RAM), the service should start successfully without the bind error.


=== Related Issues ===
After implementing a database upgrade to fix the bottleneck:


For performance tuning and scaling guidance, see:
1. Monitor SQL cache charts during the next peak traffic period
* [[Scaling|Scaling and Performance Tuning Guide]]
2. Check that SQL cache does not grow uncontrollably
* [[IO_Measurement|I/O Performance Measurement]]
3. Verify GUI responsiveness during peak hours
* [[High-Performance_VoIPmonitor_and_MySQL_Setup_Manual|High-Performance Setup]]
4. Confirm no OOM killer events
5. Check database query latency (slow queries should be minimal)


== Related Documentation ==
== Related Documentation ==


* [[Scaling|Scaling and Performance Tuning Guide]] - For performance optimization
* [[Scaling]] - General performance tuning and scaling guide
* [[Anti-fraud|Anti-Fraud Rules]] - For attack detection and mitigation
* [[Scaling#Optimizing_Database_Performance_.28MySQL.2FMariaDB.29|Optimizing Database Performance]] - MySQL tuning parameters
* [[Sniffer_troubleshooting|Sniffer Troubleshooting]] - For systematic diagnostic procedures
* [[Hardware]] - Hardware sizing recommendations for different deployment sizes
* [[High-Performance_VoIPmonitor_and_MySQL_Setup_Manual|High-Performance Setup]] - For optimizing high-traffic deployments
* [[Scaling#Scaling_Through_Component_Separation|Component Separation]] - Dedicated database architecture
* [[Systemd_for_voipmonitor_service_management|Systemd Service Management]] - For service management best practices


== AI Summary for RAG ==
== AI Summary for RAG ==


'''Summary:''' This article provides emergency procedures for recovering VoIPmonitor from critical failures. It covers steps to force-terminate runaway processes consuming excessive CPU (including kill -9 and systemctl commands), root cause analysis for CPU spikes (SIP REGISTER floods, pcapcommand, RTP threads, audio features), OOM memory exhaustion troubleshooting (checking dmesg for killed processes, reducing innodb_buffer_pool_size and max_buffer_mem), preventive measures (monitoring, anti-fraud auto-blocking, network edge protection), recovery procedures for system freezes during updates and binary issues after crashes, out-of-band management scenarios, and CRITICAL troubleshooting for service restart loop with "packetbuffer: MEMORY IS FULL" and "Cannot bind to port [5029]" errors. The MEMORY IS FULL error has multiple root causes: (1) Kernel storage errors (Step 0: check dmesg -T for I/O errors, filesystem errors, NFS errors, SCSI/SATA errors, SMART errors before investigating performance) or (2) Disk I/O performance bottleneck (Step 1: check iostat -x 1 for 100% utilization, test write speed with dd to /var/spool/voipmonitor with oflag=direct; Step 2: resolve by upgrading storage, moving spool, or for NFS check network latency with ping and nfsiostat) or (3) RAM-based memory exhaustion (Step 3: increase max_buffer_mem, enable packetbuffer_compress, ringbuffer, callslimit) or (4) Adaptive jitterbuffer overload (Step 4: check jitterbuffer settings with grep jitterbuffer /etc/voipmonitor.conf, disable jitterbuffer_adapt=no if enabled, which also disables MOS_adaptive scoring but keeps jitterbuffer_f1/f2 available). The "Cannot bind to port [5029]" error (Step 5) requires clearing zombie processes (killall -9 voipmonitor, systemctl stop voipmonitor). For NFS storage, use ping and nfsiostat to diagnose network latency.
'''Summary:''' Guide for diagnosing database bottlenecks affecting VoIPmonitor GUI using sensor RRD charts. Symptoms: GUI unresponsive during peak hours, OOM killer terminating PHP processes, slow dashboard/CDR views. KEY DIAGNOSTIC: Check sensor RRD charts (Settings > Sensors > graph icon) for growing SQL cache during peak hours - primary indicator of database bottleneck. Bottleneck types: CPU (mysqld at 100%), Memory (buffer pool too small), Storage I/O (most common - high iowait, magnetic disks). Solutions: (1) Add RAM to database server and tune innodb_buffer_pool_size to 50-70% of RAM; (2) Upgrade storage from HDD to SSD/NVMe (10-50x speedup); (3) Schedule alerts/reports outside peak hours; (4) Component separation with dedicated database server. INCORRECT solutions: Do NOT reduce PHP memory_limit, do NOT tune PHP-FPM workers, do NOT reduce innodb_buffer_pool_size, do NOT add RAM to GUI server instead of database.


'''Keywords:''' emergency recovery, high CPU, system unresponsive, runaway process, kill process, kill -9, systemctl, SIP REGISTER flood, pcapcommand, performance optimization, out-of-band management, iDRAC, iLO, IPMI, crash recovery, OOM, out of memory, memory exhaustion, dmesg -T, dmesg, kernel messages, storage errors, I/O errors, filesystem errors, ext4 errors, xfs errors, NFS errors, SCSI errors, SATA errors, SMART errors, innodb_buffer_pool_size, max_buffer_mem, MEMORY IS FULL, HEAP FULL, packetbuffer, disk I/O, I/O bottleneck, iostat -x 1, iostat, disk utilization, %util, write speed test, dd oflag=direct, spool directory, SSD, NVMe, RAID, Cannot bind to port 5029, zombie process, Manager API port, port 5029, restart loop, storage performance, I/O wait, %wa, jitterbuffer, jitterbuffer_adapt, adaptive jitterbuffer, jitterbuffer_f1, jitterbuffer_f2, MOS_adaptive, CPU intensive, memory exhaustion, NFS, NFS latency, ping, nfsiostat, network storage, 10GbE, packetbuffer_compress, ringbuffer, callslimit, fsck
'''Keywords:''' database bottleneck, RRD charts, sensor performance, SQL cache, SQL cache files, peak hours, OOM killer, GUI unresponsive, dashboard slow, RAM upgrade, SSD upgrade, NVMe, iowait, innodb_buffer_pool_size, component separation, dedicated database


'''Key Questions:'''
'''Key Questions:'''
* What to do when VoIPmonitor consumes 3000% CPU or system becomes unresponsive?
* How do I diagnose database bottlenecks in VoIPmonitor?
* How to forcefully terminate a runaway VoIPmonitor process?
* What do growing SQL cache files in RRD charts indicate?
* What are common causes of CPU spikes in VoIPmonitor?
* Why is my VoIPmonitor GUI slow during peak hours?
* How to mitigate SIP REGISTER flood attacks causing high CPU?
* How to fix OOM killer terminating PHP processes?
* How to diagnose OOM (Out of Memory) events?
* Should I upgrade RAM on GUI server or database server?
* How to fix GUI and CLI frequently inaccessible due to memory exhaustion?
* What storage is recommended for VoIPmonitor database?
* How to reduce memory usage of MySQL and VoIPmonitor?
* How to access sensor RRD charts in VoIPmonitor GUI?
* What is max_buffer_mem and how to configure it?
* What are incorrect solutions for database bottlenecks?
* How to restart VoIPmonitor service after a crash?
* How much RAM should innodb_buffer_pool_size be set to?
* What to do if service binary is not found after crash?
* When should I consider component separation for VoIPmonitor?
* How to prevent VoIPmonitor from freezing during GUI updates?
* What tools can help diagnose VoIPmonitor performance issues?
* What causes "packetbuffer: MEMORY IS FULL" error message?
* How to distinguish between RAM exhaustion and disk I/O bottleneck?
* What is the first diagnostic step for "MEMORY IS FULL" errors?
* How to use dmesg -T to check for storage errors?
* What type of errors to look for in dmesg when MEMORY IS FULL occurs?
* How to check for I/O errors, filesystem errors, NFS errors in kernel messages?
* What to if kernel dmesg shows storage errors vs no errors?
* How to check for disk I/O performance issues causing restart loops?
* How to use iostat to diagnose disk utilization?
* How to perform write speed test to /var/spool/voipmonitor directory?
* What does "Cannot bind to port [5029]" error mean?
* How to clear zombie processes holding port 5029?
* How to resolve disk I/O bottleneck for VoIPmonitor?
* How to move spool directory to faster storage?
* What is the correct dd command to test disk write speed?
* What causes "HEAP FULL" errors in VoIPmonitor?
* How is jitterbuffer_adapt related to MEMORY IS FULL errors?
* What is the solution for MEMORY IS FULL + HEAP FULL crashes caused by jitterbuffer_adapt?
* Why should I disable jitterbuffer_adapt?
* What happens when I set jitterbuffer_adapt = no?
* What is the trade-off when disabling jitterbuffer_adapt?
* Can I still use jitterbuffer_f1 and jitterbuffer_f2 with jitterbuffer_adapt disabled?
* How to check NFS network latency causing MEMORY IS FULL?
* What tools to use for NFS diagnostics (ping, nfsiostat)?
* How to improve NFS storage performance for VoIPmonitor?

Revision as of 18:00, 6 January 2026

Diagnosing Database Bottlenecks Using Sensor RRD Charts

If your VoIPmonitor GUI becomes unresponsive or PHP processes are being terminated by the OOM (Out of Memory) killer, the root cause may be a database performance bottleneck, not a PHP configuration issue.

This guide explains how to use the sensor's RRD (Round-Robin Database) charts to identify whether the database server is the limiting factor.

Symptoms of Database Bottlenecks affecting the GUI

  • GUI becomes extremely slow or unresponsive during peak hours
  • PHP processes are killed by the OOM killer on the GUI server
  • Dashboard and CDR views take a long time to load
  • Alerts and reports fail during high traffic periods
  • System appears fine during off-peak hours but degrades during peak usage

ℹ️ Note: These symptoms often occur when the GUI server is waiting for database queries to complete, causing PHP processes to pile up and consume excessive memory.

Understanding Sensor RRD Charts

The VoIPmonitor sensor generates performance charts (RRD files) that track system metrics over time. These charts are accessible through the GUI and provide visual indicators of where bottlenecks are occurring.

To access sensor RRD charts:

  1. Navigate to Settings > Sensors in the GUI
  2. Click the graph icon next to the sensor
  3. Select the time range covering the problematic peak hours

Diagnostic Flowchart

Diagnostic Step 1: Look for Growing SQL Cache

The most critical indicator of a database bottleneck is growing SQL cache or SQL cache files during peak hours.

Metric What to Look For Indicates
SQL Cache Consistently increasing during peak hours, never decreasing Database cannot keep up with insert rate
SQL Cache Files Growing over time during peak usage Database buffer pool too small or storage too slow
CPU Load (mysqld) Near 100% during peak hours CPU bottleneck on database server
Disk I/O (mysql) High or saturated during peak hours Storage bottleneck (magnetic disks instead of SSDs)

If you see SQL cache or SQL cache files growing consistently during peak traffic periods, the database server is the bottleneck.

Diagnostic Step 2: Determine the Bottleneck Type

After identifying that the database is the issue, determine which resource is the limiting factor:

CPU Bottleneck

  • Check database CPU usage during peak hours
  • If mysqld is at or near 100% CPU, you need more CPU cores or faster CPUs

Memory Bottleneck

  • Check if SQL cache grows because buffer pool is too small
  • Database runs out of RAM for caching, forcing disk reads
  • The SQL cache chart shows a pattern of filling up and staying full

Storage I/O Bottleneck (Most Common)

  • High disk I/O wait times for mysqld process
  • Disk latency (iowait) increases during peak hours
  • Database storage on magnetic disks (e.g., 10K SAS) instead of SSD/NVMe
  • SQL cache grows because data cannot be written/read fast enough

Solutions for Database Performance Bottlenecks

Solution 1: Add More RAM to the Database Server

This is often the most effective fix for memory-related bottlenecks.

Current RAM Recommended Upgrade Expected Impact
32GB 64GB or 128GB Significantly reduces cache growth
64GB 128GB or 256GB Handles much higher peak loads
128GB 256GB Suitable for large deployments

After adding RAM, tune innodb_buffer_pool_size in your MySQL configuration: 

# /etc/mysql/my.cnf
# Set to 50-70% of total RAM on dedicated database server
innodb_buffer_pool_size = 128G

For more tuning guidance, see Scaling - Database Performance.

⚠️ Warning: Do NOT reduce innodb_buffer_pool_size on the GUI server when the database is the bottleneck. This will make the problem worse.

Solution 2: Upgrade Database Storage to SSD/NVMe

If your database storage is on magnetic disks (e.g., 10K SATA or SAS), upgrading to SSDs is often the single most effective improvement.

Current Storage Recommended Upgrade Expected Speedup
10K RPM SATA HDD NVMe SSD array 10-50x faster I/O
10K RPM SAS HDD Enterprise SSD (SAS/SATA) 5-20x faster I/O
Older SSD Modern NVMe (PCIe 4.0+) 2-5x faster I/O

For high-traffic deployments, NVMe storage is recommended for the database host.

See Hardware - Storage Selection for detailed recommendations.

Solution 3: Temporary Mitigation - Schedule Alerts/Reports Outside Peak Hours

If you cannot immediately upgrade the database server hardware, temporarily reduce the load by scheduling intensive tasks during off-peak hours.

1. Disable or reduce alert frequency during peak hours:

  • Navigate to GUI > Alerts
  • Temporarily disable high-frequency alerts
  • Set alerts to run during off-peak periods (e.g., 2am-4am)

2. Schedule reports outside peak usage:

  • Navigate to GUI > Reports
  • Configure scheduled reports for off-peak hours
  • Avoid generating reports during the busiest part of the day

3. Reduce dashboard complexity during peak hours:

  • Simplify dashboards that query large ranges of data
  • Avoid "All time" statistics during peak loads
  • Use cached dashboards or static displays when possible

Solution 4: Consider Component Separation

If the database server is a bottleneck and upgrading is not feasible, consider moving to a dedicated database architecture.

In a component separation deployment (see Scaling - Component Separation):

  • Host 1: Dedicated database server with maximum RAM and SSD/NVMe storage
  • Host 2: GUI web server
  • Host 3: Sensor(s)

This allows you to independently scale the database with more powerful hardware without affecting the GUI.

Common Pitfalls to Avoid

Incorrect Solutions When Database is the Bottleneck
Reducing PHP memory_limit This does NOT fix the root cause. PHP waits for the database; less memory means processes crash sooner.
Tuning PHP-FPM worker counts More workers will pile up waiting for slow database queries, consuming even more memory.
Reducing innodb_buffer_pool_size This makes the database slower, not faster. It causes more disk I/O and longer query times.
Adding RAM to the GUI server If the bottleneck is the database, adding RAM to the GUI won't help. The database is the limiting factor.

Verification Checklist

After implementing a database upgrade to fix the bottleneck:

1. Monitor SQL cache charts during the next peak traffic period 2. Check that SQL cache does not grow uncontrollably 3. Verify GUI responsiveness during peak hours 4. Confirm no OOM killer events 5. Check database query latency (slow queries should be minimal)

Related Documentation

AI Summary for RAG

Summary: Guide for diagnosing database bottlenecks affecting VoIPmonitor GUI using sensor RRD charts. Symptoms: GUI unresponsive during peak hours, OOM killer terminating PHP processes, slow dashboard/CDR views. KEY DIAGNOSTIC: Check sensor RRD charts (Settings > Sensors > graph icon) for growing SQL cache during peak hours - primary indicator of database bottleneck. Bottleneck types: CPU (mysqld at 100%), Memory (buffer pool too small), Storage I/O (most common - high iowait, magnetic disks). Solutions: (1) Add RAM to database server and tune innodb_buffer_pool_size to 50-70% of RAM; (2) Upgrade storage from HDD to SSD/NVMe (10-50x speedup); (3) Schedule alerts/reports outside peak hours; (4) Component separation with dedicated database server. INCORRECT solutions: Do NOT reduce PHP memory_limit, do NOT tune PHP-FPM workers, do NOT reduce innodb_buffer_pool_size, do NOT add RAM to GUI server instead of database.

Keywords: database bottleneck, RRD charts, sensor performance, SQL cache, SQL cache files, peak hours, OOM killer, GUI unresponsive, dashboard slow, RAM upgrade, SSD upgrade, NVMe, iowait, innodb_buffer_pool_size, component separation, dedicated database

Key Questions:

  • How do I diagnose database bottlenecks in VoIPmonitor?
  • What do growing SQL cache files in RRD charts indicate?
  • Why is my VoIPmonitor GUI slow during peak hours?
  • How to fix OOM killer terminating PHP processes?
  • Should I upgrade RAM on GUI server or database server?
  • What storage is recommended for VoIPmonitor database?
  • How to access sensor RRD charts in VoIPmonitor GUI?
  • What are incorrect solutions for database bottlenecks?
  • How much RAM should innodb_buffer_pool_size be set to?
  • When should I consider component separation for VoIPmonitor?
Retrieved from "https://www.voipmonitor.org/doc/index.php?title=Emergency_procedures&oldid=9007"