[PYTHON] [blackbird-elasticache] Monitoring AWS ElastiCache (redis)
This Plugin is a plugin that uses the CloudWatch API to get the Metric of AWS ElastiCache. So far (I'm planning to implement memcached in the future), only redis is supported (or the project I belong to uses only redis), but Introducing blackbird-elasticache.
What metric does it get?
First, let's start with the list of Metrics that can be obtained. Statistics is, as always, how to calculate Average, Sum, Max, or Min. ElastiCache's CloudWatch is the middleware's own (in this case, redis or memcached) Metric There is a metric on the host side where the middleware is installed.
Host Side Metrics(Called Host-level metrics by Amazon)
| Metric Name | Statistics | Detail |
|---|---|---|
| CPUUtilization | Average | CPU usage. It doesn't seem to be that expensive when using KVS. |
| SwapUsage | Average | Swap usage rate. |
| FreeableMemory | Average | Available memory capacity on the host side |
| NetworkBytesIn | Average | Traffic amount for this host(in)is. |
| NetworkBytesOut | Average | Traffic amount for this host(out)is. |
about SwapUsage and FreeableMemory
Although it is said to be fully managed, I have the impression that redis is used directly for ElastiCache.
Freeable Memory and Swap Usage are items that I want to be especially careful about, Riding on the word fully managed, you may be consuming a lot of memory other than adding a cancer key. Since it is KVS, I think that Memory can be used to the limit, but it is very dangerous when swapping.
I'm sorry for what I experienced, but there was a problem that the response of redis was slow only for a specific time. As a result, I wrote too much, and only during the time when I was swapping because I swapped in at the timing of bgsave It was super late. Therefore, I think it is important to get Swap Usage on a regular basis.
Redis Metrics
| Metric Name | Statistics | Detail |
|---|---|---|
| CurrConnections | Average, Maximum | Current number of connections. max_Be careful not to hit the connections!! |
| Evictions | Average, Maximum | Number of values vibrated before the LRU deadline |
| Reclaimed | Average, Maximum | The sum of all LRUs expired and the values deleted after reaching the memory limit |
| NewConnections | Average, Maximum | Number of connections accepted within the acquisition time interval |
| BytesUsedForCached | Maximum | Amount of memory allocated by redis |
| CacheHits | Average, Maximum | Number of hits as cache |
| CacheMisses | Average, Maximum | Number of misses as cache |
| LepricationLag | Average, Maximum | Read replica delayed seconds(Only read reolica) |
| GetTypeCmds | Maximum | Total number of Get queries |
| SetTypeCmds | Maximum | Total number of Set queries |
| KeyBasedCmds | Maximum | - |
| StringBasedCmds | Maximum | - |
| HashBasedCmds | Maximum | - |
| ListBasedCmds | Maximum | - |
| SetBasedCmds | Maximum | - |
| SortedSetBasedCmds | Maximum | - |
about Evictions and Reclaimed
Evictions is the number of keys deleted before expire at the upper limit of maxmemory.
Reclaimed, on the other hand, is the total number of values deleted.
So, when the number of Reclaimed --Evictions is large, either there is not enough memory on the host side or there are a lot of useless objects.
Zabbix Template
Items
Only the values that can be obtained with the above CloudWatch API. (I'm sorry for the hurry.)
Graphs
There are the following graphs, and only the drawing of the graph that seems to be interesting is captured. (Normal line graphs and those with only one element should be omitted)
CPU Utilization
CPU usage on the host side.
Memory Usage
Freeable Memory on the host side and Used Memory on the Redis side are stacked.
Network Traffic
In and Out of the network.
Cache Hits/Miss
It is a stacked graph of Hits and Misses of Cache.
Current Items
The current number of items (a set of Key and Value).
Evictions
The number of items that have been Evictions.
Reclaimed
The number of items that are reclaimed.
New Connections
The number of Connections established per unit time.
CMDs
This is a stacked graph of the list of commands.
Summary
This plugin only takes CloudWatch Metric, but since ElastiCache is plain redis and plain memcached, blackbird-redis and [blackbird-memcached] ](Http://qiita.com/makocchi/items/d178038588465ec8ba07) may be better to see various values directly. If anything, it might be better to use both plug-ins to make it delicious (I want to do that).
Recommended Posts