Client for Throttle. Throttle is a semaphore service, providing semaphores for distributed systems.
Project description
Throttle
Semaphores for distributed systems.
Motivation
Throttle provides semaphores as a service via an http interface. As the name indicates the primary usecase in mind is to throttle a systems access to a resource, by having the elements of that system to ask for permission (i.e. acquiring a lease) first. If the system consists of several process running on different machines, or virtual machines in the same Network, throttle might fit the bill.
Throttle aims to be easy to operate, well-behaved in edge cases and works without a persistence backend.
Features
- Server builds and runs on Windows, Linux, and OS-X.
- Python Client is available.
- Fairness (longer waiting peers have priority)
- Peers acquiring locks with a large count, won't be starved by lots of others with a small one.
- Locks expire to prevent leaking semaphore count due to Network errors or client crashes.
- Locks can be prolonged indefinetly by sending heartbeats of the server.
- No persistence backend is required.
- Server recovers state from heartbeats in case of reboot.
Work in progress, interfaces are still unstable.
Usage
Operating a Throttle server
Starting and Shutdown
Assuming the throttle executable to be in your path environment variable, you start a throttle sever by executing it. You can display the availible command line options using the --help
flag. Starting it without any arguments is going to boot the server with default configuration.
throttle
This starts the server in the current process. Navigate with a browser to localhost:8000
to see a welcoming message. You can shut Throttle down gracefully by pressing Ctrl + C
.
Default logging to stderr
Set the THROTTLE_LOG
environment variable to see more output on standard error. Valid values are ERROR
, WARN
, INFO
, DEBUG
and TRACE
.
In bash:
THROTTLE_LOG=WARN
or PowerShell:
$env:THROTLLE_LOG="INFO"
Starting the server now yields more information.
[2020-04-12T18:56:23Z INFO throttle] Hello From Throttle
[2020-04-12T18:56:23Z WARN throttle] No semaphores configured.
[2020-04-12T18:56:23Z INFO actix_server::builder] Starting 8 workers
[2020-04-12T18:56:23Z INFO actix_server::builder] Starting "actix-web-service-127.0.0.1:8000" service on 127.0.0.1:8000
[2020-04-12T18:56:23Z INFO throttle::litter_collection] Start litter collection with interval: 300s
Hint: Enabling Gelf logging currently disables logging to standard error.
Toml configuration file
To actually serve semaphores, we need to configure their names and full count. By default Throttle is looking for a configuration in the working directories throttle.toml
file should it exist.
# Sample throttle.cfg Explaining the options
# The time interval in which the litter collection backgroud thread checks for expired leases.
# Default is set to 5 minutes.
litter_collection_interval = "5min"
[semaphores]
# Specify name and full count of semaphores. Below line creates a semaphore named A with a full
# count of 42. Setting the count to 1 would create a Mutex.
A = 42
# Optional logging config, to log into graylog
[logging.gelf]
name = "MyThrottleServer"
host = "my_graylog_instance.cloud"
port = 12201
# Set this to either ERROR, WARN, INFO, DEBUG or TRACE.
level = "INFO"
## Optional logging config, to log to stderr. Can be overwritten using the `THROTTLE_LOG`
## environment variable.
# [logging.stderr]
# Set this to either ERROR, WARN, INFO, DEBUG or TRACE.
# level = "INFO"
Metrics
Throttle supports Prometheus metrics, via the /metrics
endpoint. Depending on your configuration and state they may e.g. look like this:
# HELP throttle_count Accumulated count of all active leases
# TYPE throttle_count gauge
throttle_count{semaphore="A"} 0
# HELP throttle_full_count New leases which would increase the count beyond this limit are pending.
# TYPE throttle_full_count gauge
throttle_full_count{semaphore="A"} 42
# HELP throttle_longest_pending_sec Time the longest pending peer is waiting until now, to acquire a lock to a semaphore.
# TYPE throttle_longest_pending_sec gauge
throttle_longest_pending_sec{semaphore="A"} 0
# HELP throttle_num_404 Number of Get requests to unknown resource.
# TYPE throttle_num_404 counter
throttle_num_404 0
# HELP throttle_pending Accumulated count of all pending leases
# TYPE throttle_pending gauge
throttle_pending{semaphore="A"} 0
Python client
Throttle ships with a Python client. Here is how to use it in a nutshell.
from throttle_client import Peer, lock
# Configure endpoint to throttle server
p = Peer.from_server_url("http://localhost:8000")
# Use client configuraton to acquire a lock (amount 1) to semaphore A
with lock(p, "A"):
# Do stuff while holding lock to "A"
# ...
# A is released at the end of with block
Http routes
- GET
/
: Prints a greeting message - GET
/health
: Always answers with200 OK
- GET
/metrics:
: Metrics for prometheus - GET
/version
: Returns server version.
Routes for managing peers and locks
Post
new_peer
: Creates a new peer. The body to this request must contain a human readable time duration with dimension in quotes. E.g.:"expires_in": "5m"
,"expires_in": "30s"
or"expires_in": "12h"
. This is the time after which the peer is going to expire if not kept alive by prolonging its expiration time. Every lock acquired is always associated with a peer. If a peer expires, all locks are released. The request returns a random integer as peer id.Delete
/peer/{id}
: Removes the peer, releasing all its locks in the process. Every call tonew_peer
should be matched by a call to this route, so other peers do not have to wait for this peer to expire in order to acquire locks to the same semaphores.Put
/peer/{id}/{semaphore}
: Acquires lock to a semaphore for an existing peer. The body must contain the desired lock count. Throttle will answer either with200 Ok
in case the lock could be acquired, or202 Accepted
in case the lock can not be acquired until other peers release their lock. Specifying a lock count higher than the full count of the lock message or violating lock hierarchy will result in a409 Conflict
error. Requesting a lock for an unknown semaphore or unknown peer is going to result in400 Bad Request
. This request is idempotent, so acquiring locks can be repeated in case of a timeout, without risk of draining the semaphore. If waiting for a lock on the client side, busy waiting can be avoided using the optionalblock_for
query parameter. E.g./peer/{id}/{semaphore}?block_for=10s
.
WIP document routes used to acquire / relaese / hold semaphores.
Installation
Server
The server binary is published to crates.io and thus installable via cargo.
cargo install throttle-server
Python Client
Python client is publish to PyPi and can be installed using pip.
pip install throttle-client
Local build and test setup
-
Install Rust compiler and Cargo. Follow the instructions on this site
-
Checkout the sources with
git clone https://github.com/pacman82/throttle.git
-
You can run the build and run unit tests by executing
cd throttle cargo test
-
Execute integration test with clients
cd python_client pip install -r test-requirements.txt pip install -e . pytest
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for throttle_client-0.2.0-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 7006cbce883d3acfd21f3ccec1855d06dec3227569fb7dc8cb8b4a824cfeadc2 |
|
MD5 | 97b38fd956892e4f768daa779c170d39 |
|
BLAKE2b-256 | 0b3e603bc4f505079c01e0bc7a64b2d12f8d4fd860cee3ca5b53b55f9ac040a1 |