Middleware lab

The real middlewares of the client run here, in your browser, around a server that the page simulates. Change the layers and their order, inject faults into the server, and send requests. Each dot is a request, and each small dot is an item or an error on its way back.

0 sent

0 succeeded

0 failed

0 server calls

— median

— 95th percentile

Layers

From the outside in. The outside layer sees each request first and each answer last.

Cache
It keeps a successful answer for each key. A later request with the same key gets the copy and goes no deeper.
Keep an answer for5000 ms
Timeout (whole call)
It ends the call after its time, retries included, with the error TIMEOUT.
End the call after4000 ms
Retry
It starts the request again after a retryable error, with a backoff that doubles. It stops at the first item.
Retries2First backoff300 ms
Circuit breaker
After a number of failures it opens and refuses each call at once. After its reset time it lets a few probes through.
Open after failures4Probe after4000 ms
Rate limit
It lets a number of requests through in each time window. It refuses the others, or queues them.
Timeout (each attempt)
It ends one attempt after its time. A retry layer outside it can start the next attempt.

A change of the layers starts them again: the cache is empty and the breaker is closed.

The server

Latency250 msJitter100 msErrors0 % A dead peer: the server takes each request and never answersItems of an answer1

Each layer

“In” counts the requests that reached the layer. “On” counts how often it passed a request to the next layer: a retry passes one request on more than once, and a cache hit passes it on zero times.

Layer	On	Items out	Errors out
Cache	0	0	0
Timeout (whole call)	0	0	0
Retry	0	0	0
Circuit breaker (closed)	0	0	0
Server	—	the simulated server

Timeline

The last 24 requests, the newest first. A tick is one attempt at the server.

Send a request to see it here.

What happened

Nothing yet.

Middleware lab

The onion

Each layer

Timeline

What happened