SHOW WARNINGS statement can be used to retrieve the warning produced by the latest query. The error message will be returned along with the query itself:

mysql> SELECT * FROM test1 WHERE MATCH('@@title hello') \G
ERROR 1064 (42000): index test1: syntax error, unexpected TOK_FIELDLIMIT
near '@title hello'
mysql> SELECT * FROM test1 WHERE MATCH('@title -hello') \G
ERROR 1064 (42000): index test1: query is non-computable (single NOT operator)
mysql> SELECT * FROM test1 WHERE MATCH('"test doc"/3') \G
*************************** 1\. row ***************************
        id: 4
    weight: 2500
  group_id: 2
date_added: 1231721236
1 row in set, 1 warning (0.00 sec)
mysql> SHOW WARNINGS \G
*************************** 1\. row ***************************
  Level: warning
   Code: 1000
Message: quorum threshold too high (words=2, thresh=3); replacing quorum operator
         with AND operator
1 row in set (0.00 sec)

SHOW [{GLOBAL | SESSION}] VARIABLES LIKE 'pattern'

It returns the current values of a few server-wide variables. Also, support for GLOBAL and SESSION clauses was added.

mysql> SHOW GLOBAL VARIABLES;
+--------------------------+-----------+
| Variable_name            | Value     |
+--------------------------+-----------+
| autocommit               | 1         |
| collation_connection     | libc_ci   |
| query_log_format         | sphinxql  |
| log_level                | info      |
| max_allowed_packet       | 134217728 |
| character_set_client     | utf8      |
| character_set_connection | utf8      |
| grouping_in_utc          | 0         |
| last_insert_id           | 123, 200  |
+--------------------------+-----------+
9 rows in set (0.00 sec)

mysql> show variables like '%log%';
+------------------+----------+
| Variable_name    | Value    |
+------------------+----------+
| query_log_format | sphinxql |
| log_level        | info     |
+------------------+----------+
2 rows in set (0.00 sec)

mysql> show session variables like 'autocommit';
+---------------+-------+
| Variable_name | Value |
+---------------+-------+
| autocommit    | 0     |
+---------------+-------+
1 row in set (0.00 sec)

Manticore provides a Prometheus metrics endpoint and officially maintained alert rules and Grafana dashboards for Manticore Buddy.

curl -s 0:9308/metrics

# HELP manticore_uptime_seconds Time in seconds since start
# TYPE manticore_uptime_seconds counter
manticore_uptime_seconds 25
# HELP manticore_connections_count Connections count since start
# TYPE manticore_connections_count gauge
manticore_connections_count 55
# HELP manticore_maxed_out_error_count Count of maxed_out errors since start
# TYPE manticore_maxed_out_error_count counter
manticore_maxed_out_error_count 0
# HELP manticore_version Manticore Search version
# TYPE manticore_version gauge
manticore_version {version="0.0.0 c88e811b2@25060409 (columnar 5.0.1 59c7092@25060304) (secondary 5.0.1 59c7092@25060304) (knn 5.0.1 59c7092@25060304) (embeddings 1.0.0) (buddy v3.28.6-7-g14ee10)"} 1
# HELP manticore_mysql_version Manticore Search version
# TYPE manticore_mysql_version gauge
manticore_mysql_version {version="0.0.0 c88e811b2@25060409 (columnar 5.0.1 59c7092@25060304) (secondary 5.0.1 59c7092@25060304) (knn 5.0.1 59c7092@25060304) (embeddings 1.0.0)"} 1
# HELP manticore_command_search_count Count of search queries since start
# TYPE manticore_command_search_count counter
manticore_command_search_count 1
......

This folder contains officially maintained monitoring assets for Manticore Search.

File: manticore-alerts.yml

Example scrape config:

scrape_configs:
  - job_name: manticoresearch
    metrics_path: /metrics
    static_configs:
      - targets:
          - 127.0.0.1:9308

• ManticoreBuddyTargetDown: Prometheus cannot scrape this target. Means the service is down, unreachable, or the scrape config is wrong.
• ManticoreBuddyRecentlyRestarted: Uptime stays under 5 minutes for 5 minutes. Means the process is restarting or flapping.
• ManticoreBuddyMaxedOutErrors: The maxed_out error counter increases. Means Manticore is rejecting work due to limits.
• ManticoreBuddySearchLatencyP95High: p95 search latency stays above 500 ms for 10 minutes. Most users feel slow searches.
• ManticoreBuddySearchLatencyP99High: p99 search latency stays above 1000 ms for 10 minutes. The slowest requests are very slow.
• ManticoreBuddyWorkQueueBacklog: Work queue length stays above 100 for 5 minutes. Requests are piling up.
• ManticoreBuddyWorkersSaturated: More than 90% of workers are active for 10 minutes. Threads are maxed, so latency will rise.
• ManticoreBuddyQueryCacheNearLimit: Query cache stays above 90% usage for 10 minutes. Cache churn likely, expect slower queries.
• ManticoreBuddyDiskMappedCacheLow: Disk mapped cache ratio stays below 50% for 15 minutes. More disk IO, slower queries.
• ManticoreBuddyDiskMappedCacheVeryLow: Disk mapped cache ratio stays below 20% for 15 minutes. Severe disk IO, high latency risk.

• Availability and restarts catch hard outages and flapping early.
• Latency and queueing detect overload before errors appear.
• Cache and disk-mapped ratio are early warnings for IO-bound slowdowns.
• Worker saturation and maxed_out errors indicate capacity limits are being hit.

• TargetDown: check service health, network, scrape config, and firewall.
• RecentlyRestarted: check logs, crashes, OOM, or restarts from orchestration.
• MaxedOutErrors: check resource limits, worker count, and query concurrency.
• Latency/Queue/Saturation: reduce load, scale resources, or tune queries/indexes.
• Cache alerts: increase cache size or reduce cache-churning queries.
• Disk mapped cache ratio alerts: add RAM, reduce working set, or optimize tables.

File: manticore-buddy-dashboard.json

• Uptime: Time since the last restart. If this keeps dropping, the service is restarting or crashing.
• Version: The exact Manticore build running. If behavior changes after a deploy, check this first.
• Current Connections: How many client connections are open right now. A sudden spike = traffic surge; a sudden drop = outage or client issues.
• Active Workers: How many worker threads are busy. If this is near the total for a long time, the server is overloaded.
• Load (All Queues): Average queue load over 1/5/15 minutes. If all three lines climb, the system is falling behind.
• Work Queue Length: How many tasks are waiting. If this keeps growing, requests are piling up and latency will rise.
• Commands per Second: How many searches/inserts/updates/etc. per second. Tells you if traffic is heavy and what kind.
• Search Latency (p95/p99): 95%/99% of searches are faster than this. If these grow, users feel slow searches even if averages look fine.
• Query Cache Usage: How much cache is used vs the max. If usage keeps growing to the limit, the cache will churn and queries slow down.
• Query Cache Hit Rate: How many cache hits per second. If this goes down while traffic stays high, the cache is not helping.
• Disk and RAM Bytes: Size of indexed data on disk and RAM. Rapid growth means more storage pressure and usually slower queries.
• Disk Mapped Cache Ratio by Table: How much of each table is already cached in memory. Low ratios mean more disk reads and slower queries.
• Slowest Thread: How long the slowest running query takes. If this spikes, there may be stuck or very heavy queries.
• Connections by Type: Splits connections into current, buddy, and vip. Helpful to see which protocol is driving load.

1. Open Grafana and go to Dashboards -> Import.
2. Upload the JSON file.
3. Select your Prometheus datasource when prompted.