Name: akubra
Owner: Allegro Tech
Description: Simple solution to keep a independent S3 storages in sync
Created: 2016-10-24 11:32:11.0
Updated: 2018-04-24 21:54:35.0
Pushed: 2018-05-24 13:43:41.0
Size: 8987
Language: Go
GitHub Committers
User | Most Recent Commit | # Commits |
---|
Other Committers
User | Most Recent Commit | # Commits |
---|
Akubra is a simple solution to keep an independent S3 storages in sync - almost realtime, eventually consistent.
Keeping synchronized storage clusters, which handles great volume of new objects (about 300k obj/h), is the most efficient by feeding them with all incoming data at once. That's what Akubra does, with a minimum memory and cpu footprint.
Synchronizing S3 storages offline is almost impossible with a high volume traffic. It would require keeping track of new objects (or periodical bucket listing), downloading and uploading them to other storage. It's slow, expensive and hard to implement robustly.
Akubra way is to put files in all storages at once by copying requests to multiple backends. I case one of clusters rejects request it logs that event, and syncronizes troublesome object with an independent process.
Akubra has sharding capabilities. You may easily configure new backends with weigths and append them to regions cluster pool.
Based on clusters weights akubra splits all operations between clusters in pool. It also backtracks to older cluster when requested for not existing object on target cluster. This kind of events are logged, so it's possible to rebalance clusters in background.
You need go >= 1.8 compiler see
In main directory of this repository do:
build
test
e: akubra [<flags>]
s:
--help Show context-sensitive help (also try --help-long and --help-man).
, --conf=CONF Configuration file e.g.: "conf/dev.yaml"
ra -c devel.yaml
Once a request comes to our proxy we copy all its headers and create pipes for body streaming to each endpoint. If any endpoint returns a positive response it's immediately returned to a client. If all endpoints return an error, then the first response is passed to the client
If some nodes respond incorrectly we log which cluster has a problem, is it storing or reading and where the erroneous file may be found. In that case we also return positive response as stated above.
We also handle slow endpoint scenario. If there are more connections than safe limit defined in configuration, the backend with most of them is taken out of the pool and error is logged.
Configuration is read from a YAML configuration file with the following fields:
sten interface and port e.g. "127.0.0.1:9090", ":80"
en: ":8080"
chnical endpoint interface
nicalEndpointListen: ":8071"
chnical health check endpoint (for load balancers)
thCheckEndpoint: "/status/ping"
ditional not AWS S3 specific headers proxy will add to original request
tionalRequestHeaders:
'Cache-Control': "public, s-maxage=600, max-age=600"
'X-Akubra-Version': '0.9.26'
ditional headers added to backend response
tionalResponseHeaders:
'Access-Control-Allow-Origin': "*"
'Access-Control-Allow-Credentials': "true"
'Access-Control-Allow-Methods': "GET, POST, OPTIONS"
'Access-Control-Allow-Headers': "DNT,X-CustomHeader,Keep-Alive,User-Agent,X-Requested-With,If-Modified-Since,Cache-Control,Content-Type"
xIdleConns see: https://golang.org/pkg/net/http/#Transport
fault 0 (no limit)
dleConns: 0
xIdleConnsPerHost see: https://golang.org/pkg/net/http/#Transport
fault 100
dleConnsPerHost: 100
leConnTimeout see: https://golang.org/pkg/net/http/#Transport
fault 0 (no limit)
ConnTimeout: 0s
sponseHeaderTimeout see: https://golang.org/pkg/net/http/#Transport
fault 5s
onseHeaderTimeout: 5s
sableKeepAlives see: https://golang.org/pkg/net/http/#Transport
fault false
bleKeepAlives: false
ximum accepted body size
MaxSize: "100M"
ximum number of incoming requests to process at once
oncurrentRequests: 200
ckend in maintenance mode. Akubra will skip this endpoint
intainedBackends:
"http://s3.dc2.internal"
st request methods to be logged in synclog in case of backend failure
LogMethods:
PUT
DELETE
nfigure sharding
ters:
uster1:
Backends:
- http://127.0.0.1:9001
uster2:
Backends:
- http://127.0.0.1:9002
ons:
region:
Clusters:
- Cluster: cluster1
Weight: 0
- Cluster: cluster2
Weight: 1
Domains:
- myregion.internal
ing:
nclog:
stderr: true
stdout: false # default: false
file: "/var/log/akubra/sync.log" # default: ""
syslog: LOG_LOCAL1 # default: LOG_LOCAL1
database:
user: dbUser
password: ""
dbname: dbName
host: localhost
inserttmpl: |
INSERT INTO tablename(path, successhost, failedhost, ts,
method, useragent, error)
VALUES ('new','{{.path}}','{{.successhost}}','{{.failedhost}}',
'{{.ts}}'::timestamp, '{{.method}}','{{.useragent}}','{{.error}}');
inlog:
stderr: true
stdout: false # default: false
file: "/var/log/akubra/akubra.log" # default: ""
syslog: LOG_LOCAL2 # default: LOG_LOCAL2
level: Error # default: Debug
cesslog:
stderr: true # default: false
stdout: false # default: false
file: "/var/log/akubra/access.log" # default: ""
syslog: LOG_LOCAL3 # default: LOG_LOCAL3
able metrics collection
ics:
Possible targets: "graphite", "expvar", "stdout"
rget: graphite
Expvar handler listener address
pAddr: ":8080"
How often metrics should be released, applicable for "graphite" and "stdout"
terval: 30s
Graphite metrics prefix path
efix: my.metrics
Shall prefix be suffixed with "<hostname>.<process>"
pendDefaults: true
Graphite collector address
dr: graphite.addr.internal:2003
Debug includes runtime.MemStats metrics
bug: false
Akubra has technical http endpoint for configuration validation puroposes. It's configured with TechnicalEndpointListen property.
curl -vv -X POST -H "Content-Type: application/yaml" --data-binary @akubra.cfg.yaml http://127.0.0.1:8071/validate/configuration
Possible responses:
* HTTP 200
Configuration checked - OK.
or:
* HTTP 400, 405, 413, 415 and info in body with validation error message
Feature required by load balancers, DNS servers and related systems for health checking.
In configuration YAML we have parameter HealthCheckEndpoint
- it's URI path for health check HTTP endpoint.
curl -vv -X GET http://127.0.0.1:8080/status/ping
Response:
< HTTP/1.1 200 OK
< Cache-Control: no-cache, no-store
< Content-Type: text/html
< Content-Length: 2
OK