andyxning/shortme: Yet Another URL Shortening Service in Golang · GitHub

ShortMe is a url shortening service written in Golang.
It is with high performance and scalable.
ShortMe is ready to be used in production. Have fun with it. :)

• Convert same long urls to different short urls.
• Api support
• Web support
• Short url black list
  • To avoid some words, like f**k and stupid
  • To make sure that apis such as /version and /health will only be used as api not short urls or otherwise when requesting http://127.0.0.1:3030/version, version info will be returned rather the long url corresponding to the short url "version".
• Base string config in configuration file
  • Once this base string is specified, it can not be reconfigured anymore otherwise the shortened urls may not be unique and thus may conflict with previous ones.
• Avoid short url loop
  • In case we request the short url for an already shortened url by shortme. This is meaningless and will consume more resource in shortme.
• Short http or https urls

Currently, afaik, there are three ways to implement short url service.

• Hash

  • This way is straightforward. However, every hash function will have a collision when data is large.

• Sample

  • this way may contain collision, too. See example below (This example, in Python, is only used to demonstrate the collision situation.).

  >>> import random
  >>> import string
  >>> random.sample('abc', 2)
  ['c', 'a']
  >>> random.sample('abc', 2)
  ['a', 'b']
  >>> random.sample('abc', 2)
  ['c', 'b']
  >>> random.sample('abc', 2)
  ['a', 'b']
  >>> random.sample('abc', 2)
  ['b', 'c']
  >>> random.sample('abc', 2)
  ['b', 'c']
  >>> random.sample('abc', 2)
  ['c', 'a']
  >>>

• Base

  • Just like converting bytes to base64 ascii, we can convert base10 to base62 and then make a map between 0 .. 61 to a-zA-Z0-9. At last, we can get a unique string if we can make sure that the integer is unique. So, the URL shortening question transforms into making sure we can get a unique integer. ShortMe Use the method that Flicker use to generate a unique integer(Auto_increment + Replace into + MyISAM). Currently, we only use one backend db to generate sequence. For multiple sequence counter db configuration see [Deploy#Sequence Database] (#Sequence Database)

• /version
  • HTTP GET
  • Version info
  • Example
    • curl http://127.0.0.1:3030/version
• /health
  • HTTP GET
  • Health check
  • Example
    • curl http://127.0.0.1:3030/health
• /short
  • HTTP POST
  • Short the long url
  • Example
    • curl -X POST -H "Content-Type:application/json" -d "{\"longURL\": \"http://www.google.com\"}" http://127.0.0.1:3030/short
• /{a-zA-Z0-9}{1,11}

  • HTTP GET

  • Expand the short url and return a temporary redirect HTTP status

  • Example

    • curl -v http://127.0.0.1:3030/3

        *   Trying 127.0.0.1...
        * Connected to 127.0.0.1 (127.0.0.1) port 3030 (#0)
        > GET /3 HTTP/1.1
        > Host: 127.0.0.1:3030
        > User-Agent: curl/7.43.0
        > Accept: */*
        >
        < HTTP/1.1 307 Temporary Redirect
        < Location: http://www.google.com
        < Date: Fri, 15 Apr 2016 07:25:24 GMT
        < Content-Length: 0
        < Content-Type: text/plain; charset=utf-8
        <
        * Connection #0 to host 127.0.0.1 left intact

The web interface mainly used to make url shorting service more intuitively.

For short option, the shorted url, shorted url qr code and the corresponding long page is shown.

For expand option, the expanded url, expanded url qr code and the corresponding expanded page is shown.

Projects that use short-me.

• short-url

• Golang
• Mysql

mkdir -p $GOPATH/src/github.com/andyxning
cd $GOPATH/src/github.com/andyxning
git clone https://github.com/andyxning/shortme.git

cd shortme
make build

We use two databases. Import the two schemas.

• shortme
  • Store short url info
  • shortme schema
• sequence
  • sequence generator
  • sequence schema

[http]
# Listen address
listen = "0.0.0.0:3030"

[sequence_db]
# Mysql sequence generator DSN
dsn = "sequence:sequence@tcp(127.0.0.1:3306)/sequence"

# Mysql connection pool max idle connection
max_idle_conns = 4

# Mysql connection pool max open connection
max_open_conns = 4

[short_db]
# Mysql short service read db DSN
read_dsn = "shortme_w:shortme_w@tcp(127.0.0.1:3306)/shortme"

# Mysql short service write db DSN
write_dsn = "shortme_r:shortme_r@tcp(127.0.0.1:3306)/shortme"

# Mysql connection pool max idle connection
max_idle_conns = 8

# Mysql connection pool max open connection
max_open_conns = 8

[common]
# short urls that will be filtered to use
black_short_urls = ["version","health","short","expand","css","js","fuck","stupid"]

# Base string used to generate short url
base_string = "Ds3K9ZNvWmHcakr1oPnxh4qpMEzAye8wX5IdJ2LFujUgtC07lOTb6GYBQViSfR"

# Short url service domain name. This is used to filter short url loop.
domain_name = "short.me:3030"

# Short url service schema: http or https.
schema = "http"

We use an Mysql unsigned bigint type to store the sequence counter. According to the Mysql doc we can get 18446744073709551616 different integers. However, according to Golang doc about LastInsertId the returned auto increment integer can only be int64 which will make the sequence smaller than uint64. Even through, we can still get 9223372036854775808 different integers and this will be large enough for most service.

Supposing that we consume 100,000,000 short urls one day, then the sequence counter can last for 2 ** 63 / 100000000 / 365 = 252695124 years.

The max string length needed for encoding 2 ** 63 integers will be 11.

>>> 62 ** 10
839299365868340224
>>> 2 ** 63
9223372036854775808L
>>> 62 ** 11
52036560683837093888L

After setting up the databases and before running shortme, make sure that the corresponding user and password has been granted. After logging in mysql console, run following sql statement:

• grant insert, delete on sequence.* to 'sequence'@'%' identified by 'sequence'
• grant insert on shortme.* to 'shortme_w'@'%' identified by 'shortme_w'
• grant select on shortme.* to 'shortme_r'@'%' identified by 'shortme_r'

• make sure that static directory will be at the same directory as shortme
• ./shortme -c config.conf

In the Flickr blog, Flickr suggests that we can use two databases with one for even sequence and the other one for odd sequence. This will make sequence generator being more available in case one database is down and will also spread the load about generate sequence. After splitting sequence db from one to more, we can use HaProxy as a reverse proxy and thus more sequence databases can be used as one. As for load balance algorithm, i think round robin is good enough for this situation.

In two databases situation, we should add the following configuration to each database configuration file.

• First database

auto_increment_offset 1
auto_increment_increment 2

• Second databse

auto_increment_offset 2
auto_increment_increment 2

Then each time to generate a sequence counter, we can execute below sql statement:
replace into sequence(stub) values("sequence")

In cases we use three databases as sequence counter generator, we should insert a record for each table in two databases.

• First database

auto_increment_offset 1
auto_increment_increment 3

• Second database

auto_increment_offset 2
auto_increment_increment 3

• Third database

auto_increment_offset 3
auto_increment_increment 3

Then each time to generate a sequence counter, we can execute below sql statement:
replace into sequence(stub) values("sequence")

Ok, i think you get the point. When using N databases to generate sequence counter, configuration for each database configuration file will just like below:

for i := range N {
    add "auto_increment_offset i" to config file
    add "auto_increment_increment N" to config file
}

So, sequence generator can be horizontally scalable.

With short urls increasing, many records are stored in one table. This is not an optimal mysql practice. In this case we can simply shard table to bypass this problem.

For example, we can shard according to the base integer using modula hash algorithm. This has a good distribution between tables. We can use 100 short tables with names like short_00/short_01/short_02/.. ./short_99. we can use pseudo code blow to determine which is the table to store the short url record.

baseInteger := sequence.NextSequence()
tableName := fmt.Sprintf("short_%s", baseInteger % 100)

There are many table sharding algorithms, we can shard table according to range id, user name and so on. If we use user name as the criteria to shard table, we can do some aggregate algorithm like how many records a user has created easily. This may also has some drawbacks such as if user Lily and user Lucy are sharded to different tables and Lily shorts about 1k urls Lucy shorts about 1M urls, then we may encounter the unbalance hash problem, i.e., some tables contains more records than others.

In conclusion, there are many factors to consider before we can make a decision which hash algorithm to use.

Sometimes we may want to make some statistics about hit number, UA(User Agent), original IP and so on.

A recommended way to deploy shortme is to use it behind a reverse proxy server such as Nginx. Under this way, the statistics info can be analysed by analysing the access log of Nginx. This is can be accomplished by awk or more trending log analyse stack ELK.

• long url may make the generated qr code unreadable. I have test this in my self phone. This remains to be tested more meticulous.
• One demand about customizing the short url can not be done easily currently in shortme according to the id generation logic. Let's make it happen. :)