[redis] values transformed into tags in prometheus output

[axinojolais]

Hi,

Using telegraf-1.8.3, redis version 4.0.11.

Some values reported by the redis plugin are considered as tags by the prometheus output, presumably because they're strings and not numerals. As you can see in the test below, the tags are host, port, replication_role and server :

$ telegraf --config /etc/telegraf/telegraf.d/redis.conf --input-filter redis --test
> redis,host=<foo>,port=6379,replication_role=master,server=1.2.3.4 active_defrag_hits=0i,active_defrag_key_hits=0i,active_defrag_key_misses=0i,active_defrag_misses=0i,active_defrag_running=0i,aof_current_rewrite_time_sec=-1i,aof_enabled=0i,aof_last_bgrewrite_status="ok",aof_last_cow_size=0i,aof_last_rewrite_time_sec=-1i,aof_last_write_status="ok",aof_rewrite_in_progress=0i,aof_rewrite_scheduled=0i,blocked_clients=0i,client_biggest_input_buf=0i,client_longest_output_list=0i,clients=2i,cluster_enabled=0i,connected_slaves=0i,evicted_keys=0i,expired_keys=0i,expired_stale_perc=0,expired_time_cap_reached_count=0i,instantaneous_input_kbps=0,instantaneous_ops_per_sec=0i,instantaneous_output_kbps=0,keyspace_hitrate=0,keyspace_hits=0i,keyspace_misses=0i,latest_fork_usec=0i,lazyfree_pending_objects=0i,loading=0i,lru_clock=16504280i,master_repl_offset=0i,maxmemory=0i,maxmemory_policy="noeviction",mem_fragmentation_ratio=6.94,migrate_cached_sockets=0i,pubsub_channels=0i,pubsub_patterns=0i,rdb_bgsave_in_progress=0i,rdb_changes_since_last_save=0i,rdb_current_bgsave_time_sec=-1i,rdb_last_bgsave_status="ok",rdb_last_bgsave_time_sec=-1i,rdb_last_cow_size=0i,rdb_last_save_time=1543225250i,rdb_last_save_time_elapsed=5686i,redis_version="4.0.11",rejected_connections=0i,repl_backlog_active=0i,repl_backlog_first_byte_offset=0i,repl_backlog_histlen=0i,repl_backlog_size=1048576i,second_repl_offset=-1i,slave_expires_tracked_keys=0i,sync_full=0i,sync_partial_err=0i,sync_partial_ok=0i,total_commands_processed=346i,total_connections_received=73i,total_net_input_bytes=4877i,total_net_output_bytes=974798i,total_system_memory=4136558592i,uptime=5686i,used_cpu_sys=7.17,used_cpu_sys_children=0,used_cpu_user=2.79,used_cpu_user_children=0,used_memory=590976i,used_memory_dataset=16384i,used_memory_dataset_perc="19.77%",used_memory_lua=37888i,used_memory_overhead=574592i,used_memory_peak=590976i,used_memory_peak_perc="100.00%",used_memory_rss=4104192i,used_memory_startup=508104i 1543230936000000000

However, if I curl telegraf, a few of these are switched to tags :

$ curl -s localhost:9103/metrics|egrep ^redis_used_cpu_user_children
redis_used_cpu_user_children{aof_last_bgrewrite_status="ok",aof_last_write_status="ok",host="<foo>",maxmemory_policy="noeviction",port="6379",rdb_last_bgsave_status="ok",redis_version="4.0.11",replication_role="master",server="1.2.3.4",used_memory_dataset_perc="19.96%",used_memory_peak_perc="96.47%"} 0

I don't mind redis_version, but the rest doesn't belong as tags. Transforming the percentages into numbers should be easy (just drop the %), but for the rest I don't have a good solution to propose.

Thanks

[glinton]

Isn't that just how prometheus works, storing only key/value pairs? If you want to preserve the fields and tags influxdb would be a better choice as a backend.

[danielnelson]

I think you probably have string_as_label set to true, if you disable it should be fixed. Let us know if this helps.

[axinojolais]

Indeed, disabling string_as_label works. Thanks !