Generating YouTube-like IDs in Postgres using PL/V8 and Hashids
Recently on a Rails project, I ran into an issue where I wanted to expose a resource (lets say it was a product) in a RESTful route, but I also didn’t want the URLs to be easily guessable. In other words, following Rails conventions my standard “show” actions would be URLs like https://example.com/products/1, https://example.com/products/2, https://example.com/products/3, which are trivially guessable since we’re exposing the database’s auto-incrementing integer primary key as the resource ID. To prevent people from writing a super simple script that could scrape my whole product catalog, it would be nice if we could make the URLs not trivially guessable while still remaining publicly-accessable for people who know them.
One approach that some people advocate is simply using UUIDs, but I think URLs like https://example.com/products/3bc95fb9-f0c1-4af8-989e-6ea8467879d3 simply look nasty, particularly when you get into nested sub-resources with their own UUIDs tacked on. It’s something I don’t want to subject my users’ eyes to or have potentially affect SEO / page rank due to the extraneous length.1
Hashids
A nice compromise here is using a library called Hashids, which can take an integer input (e.g. our primary keys), and a salt, and obfuscate2 them into YouTube-like, short, non-guessable IDs like these: https://example.com/products/NV, https://example.com/products/6m, https://example.com/products/yD.
The Hashids project links to many implementations and documentation in various languages, including Ruby. Since my project is using Rails, a simple solution would be to add an after_create callback to my model to set an attribute using the Ruby library:
# == Schema Information
#
# Table name: products
#
# id :integer not null, primary key
# title :string
# hashid :string
#
# Indexes
#
# index_products_on_hashid (hashid)
#
class Product < ActiveRecord::Base
after_create :save_hashid
private
def save_hashid
unless self.hashid
h = Hashids.new(ENV["HASHID_SALT"], ENV["HASHID_MIN_LENGTH"].to_i)
self.update!(hashid: h.encode(self.id))
end
end
end
This works! However there are at least two drawbacks:
- Creating a
Productrequires two round-trips to the database: an INSERT to create the record with a NULL value in thehashidcolumn, then an UPDATE after Rails gets the value of the integeridcolumn and can calculate the Hashid value, and update the record with it. This should be safe in terms of not leaving half-bakedproductsrecords with NULLhashidvalues out there, since Rails runsafter_createcallbacks in the same transaction that creates the record, but it’s not good performance-wise. - Somewhat related to the first drawback, the schema for this table is not optimal as the
hashidcolumn should really have a NOT NULL constraint with a UNIQUE index. But using Rails callbacks forces it to be this way. It would be much more preferable if we could lean on the database to enforce referential integrity; at my job we’ve seen plenty of instances of bad data getting into loose schemas that Should Never Happen™ from the application’s point of view.
If only there were a way for Postgres to populate that column instead…
Executing JavaScript in Postgres using PL/V8
Luckily there is a way to do this using a Postgres extension that embeds the V8 JavaScript engine in Postgres called PL/V8!3
On Ubuntu, installing PL/V8 is as easy as doing sudo apt-get install postgresql-9.6-plv8 (substitute 9.6 with whatever Postgres version you have installed) and restarting the database cluster with sudo service postgres restart. Then, open a SQL prompt on the database you want to enable it for, and execute CREATE EXTENSION plv8;. Now you can write JavaScript functions in the database!
The first step is writing a function to load the Hashids library:
| CREATE OR REPLACE FUNCTION load_hashids() RETURNS VOID AS $$ | |
| (function() { | |
| !function(t,e){if("function"==typeof define&&define.amd)define(["module","exports"],e);else if("undefined"!=typeof exports)e(module,exports);else{var s={exports:{}};e(s,s.exports),t.Hashids=s.exports}}(this,function(t,e){"use strict";function s(t,e){if(!(t instanceof e))throw new TypeError("Cannot call a class as a function")}Object.defineProperty(e,"__esModule",{value:!0});var h=function(){function t(t,e){for(var s=0;s<e.length;s++){var h=e[s];h.enumerable=h.enumerable||!1,h.configurable=!0,"value"in h&&(h.writable=!0),Object.defineProperty(t,h.key,h)}}return function(e,s,h){return s&&t(e.prototype,s),h&&t(e,h),e}}(),r=function(){function t(){var e=arguments.length<=0||void 0===arguments[0]?"":arguments[0],h=arguments.length<=1||void 0===arguments[1]?0:arguments[1],r=arguments.length<=2||void 0===arguments[2]?"abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890":arguments[2];s(this,t);var a=16,n=3.5,i=12,l="error: alphabet must contain at least X unique characters",u="error: alphabet cannot contain spaces",p="",o=void 0,f=void 0;this.escapeRegExp=function(t){return t.replace(/[-[\]{}()*+?.,\\^$|#\s]/g,"\\$&")},this.parseInt=function(t,e){return/^(\-|\+)?([0-9]+|Infinity)$/.test(t)?parseInt(t,e):NaN},this.seps="cfhistuCFHISTU",this.minLength=parseInt(h,10)>0?h:0,this.salt="string"==typeof e?e:"","string"==typeof r&&(this.alphabet=r);for(var g=0;g!==this.alphabet.length;g++)p.indexOf(this.alphabet.charAt(g))===-1&&(p+=this.alphabet.charAt(g));if(this.alphabet=p,this.alphabet.length<a)throw l.replace("X",a);if(this.alphabet.search(" ")!==-1)throw u;for(var c=0;c!==this.seps.length;c++){var b=this.alphabet.indexOf(this.seps.charAt(c));b===-1?this.seps=this.seps.substr(0,c)+" "+this.seps.substr(c+1):this.alphabet=this.alphabet.substr(0,b)+" "+this.alphabet.substr(b+1)}this.alphabet=this.alphabet.replace(/ /g,""),this.seps=this.seps.replace(/ /g,""),this.seps=this._shuffle(this.seps,this.salt),(!this.seps.length||this.alphabet.length/this.seps.length>n)&&(o=Math.ceil(this.alphabet.length/n),o>this.seps.length&&(f=o-this.seps.length,this.seps+=this.alphabet.substr(0,f),this.alphabet=this.alphabet.substr(f))),this.alphabet=this._shuffle(this.alphabet,this.salt);var d=Math.ceil(this.alphabet.length/i);this.alphabet.length<3?(this.guards=this.seps.substr(0,d),this.seps=this.seps.substr(d)):(this.guards=this.alphabet.substr(0,d),this.alphabet=this.alphabet.substr(d))}return h(t,[{key:"encode",value:function(){for(var t=arguments.length,e=Array(t),s=0;s<t;s++)e[s]=arguments[s];var h="";if(!e.length)return h;if(e[0]&&e[0].constructor===Array&&(e=e[0],!e.length))return h;for(var r=0;r!==e.length;r++)if(e[r]=this.parseInt(e[r],10),!(e[r]>=0))return h;return this._encode(e)}},{key:"decode",value:function(t){var e=[];return t&&t.length&&"string"==typeof t?this._decode(t,this.alphabet):e}},{key:"encodeHex",value:function(t){if(t=t.toString(),!/^[0-9a-fA-F]+$/.test(t))return"";for(var e=t.match(/[\w\W]{1,12}/g),s=0;s!==e.length;s++)e[s]=parseInt("1"+e[s],16);return this.encode.apply(this,e)}},{key:"decodeHex",value:function(t){for(var e=[],s=this.decode(t),h=0;h!==s.length;h++)e+=s[h].toString(16).substr(1);return e}},{key:"_encode",value:function(t){for(var e=void 0,s=this.alphabet,h=0,r=0;r!==t.length;r++)h+=t[r]%(r+100);e=s.charAt(h%s.length);for(var a=e,n=0;n!==t.length;n++){var i=t[n],l=a+this.salt+s;s=this._shuffle(s,l.substr(0,s.length));var u=this._toAlphabet(i,s);if(e+=u,n+1<t.length){i%=u.charCodeAt(0)+n;var p=i%this.seps.length;e+=this.seps.charAt(p)}}if(e.length<this.minLength){var o=(h+e[0].charCodeAt(0))%this.guards.length,f=this.guards[o];e=f+e,e.length<this.minLength&&(o=(h+e[2].charCodeAt(0))%this.guards.length,f=this.guards[o],e+=f)}for(var g=parseInt(s.length/2,10);e.length<this.minLength;){s=this._shuffle(s,s),e=s.substr(g)+e+s.substr(0,g);var c=e.length-this.minLength;c>0&&(e=e.substr(c/2,this.minLength))}return e}},{key:"_decode",value:function(t,e){var s=[],h=0,r=new RegExp("["+this.escapeRegExp(this.guards)+"]","g"),a=t.replace(r," "),n=a.split(" ");if(3!==n.length&&2!==n.length||(h=1),a=n[h],"undefined"!=typeof a[0]){var i=a[0];a=a.substr(1),r=new RegExp("["+this.escapeRegExp(this.seps)+"]","g"),a=a.replace(r," "),n=a.split(" ");for(var l=0;l!==n.length;l++){var u=n[l],p=i+this.salt+e;e=this._shuffle(e,p.substr(0,e.length)),s.push(this._fromAlphabet(u,e))}this._encode(s)!==t&&(s=[])}return s}},{key:"_shuffle",value:function(t,e){var s=void 0;if(!e.length)return t;for(var h=t.length-1,r=0,a=0,n=0;h>0;h--,r++){r%=e.length,a+=s=e.charAt(r).charCodeAt(0),n=(s+r+a)%h;var i=t[n];t=t.substr(0,n)+t.charAt(h)+t.substr(n+1),t=t.substr(0,h)+i+t.substr(h+1)}return t}},{key:"_toAlphabet",value:function(t,e){var s="";do s=e.charAt(t%e.length)+s,t=parseInt(t/e.length,10);while(t);return s}},{key:"_fromAlphabet",value:function(t,e){for(var s=0,h=0;h<t.length;h++){var r=e.indexOf(t[h]);s+=r*Math.pow(e.length,t.length-h-1)}return s}}]),t}();e.default=r,t.exports=e.default}); | |
| })() | |
| $$ LANGUAGE plv8 IMMUTABLE STRICT; |
(The above is simply the source for hashids.min.js wrapped in an immediately-executed anonymous function).
After executing that DDL to create the function, execute this SQL to run it:
SELECT load_hashids();
And now, the Hashids constant is ready for use in any JavaScript code inside PL/V8 functions for the remainder of the SQL session (each session gets its own global JS runtime context). We can now do a quick test of the Hashids library inside Postgres:
DO LANGUAGE PLV8 $$
var h = new Hashids('foo');
plv8.elog(NOTICE,h.encode(123));
$$;
You should see NOTICE: 1yR in the output, confirming it works!
As mentioned, this constant will only live as long as the SQL session. A new connection will require rerunning SELECT load_hashids(); to make it available again. Luckily, PL/V8 comes with support for a postgresql.conf configuration value we can use to load a custom PL/V8 function when the runtime is initialized. Simply add this to to postgresql.conf:
plv8.start_proc = 'load_hashids'
And now that is all handled for us!
An example usage
Now let’s put it all together with an example that fixes my issue with products. First, let’s make a helper SQL function to generate Hashids that we’ll be able to call from other SQL functions (like triggers):
CREATE FUNCTION gen_hashid(salt TEXT, min_length BIGINT, key BIGINT) RETURNS TEXT AS $$
var h = new Hashids(salt, min_length);
return h.encode(key);
$$ LANGUAGE PLV8 IMMUTABLE STRICT;
This can be tested like so:
SELECT gen_hashid('foo', 5, 123);
Which should output 61yR6.
Next, here’s a little mockup of a products schema that uses a pre-insert trigger to automatically generate Hashids:
| CREATE TABLE products ( | |
| id BIGSERIAL, | |
| title TEXT NOT NULL, | |
| hashid TEXT NOT NULL | |
| ); | |
| CREATE FUNCTION products_pre_insert() RETURNS trigger AS $$ | |
| BEGIN | |
| NEW.hashid := gen_hashid('products_secret_salt_here', 3, NEW.id); | |
| RETURN NEW; | |
| END; | |
| $$ LANGUAGE plpgsql; | |
| CREATE TRIGGER products_pre_insert BEFORE INSERT ON products FOR EACH ROW EXECUTE PROCEDURE products_pre_insert(); |
Now let’s test it out by inserting some test records:
INSERT INTO products (title) VALUES ('foo');
INSERT INTO products (title) VALUES ('foo');
INSERT INTO products (title) VALUES ('bar');
INSERT INTO products (title) VALUES ('baz');
And now let’s see what SELECT * FROM products returns:
id | title | hashid
----+-------+--------
1 | foo | WmX
2 | foo | 4zq
3 | bar | eJk
4 | baz | eEp
(4 rows)
Works beautifully! My problem is solved.
Note that in this example I hardcoded the salt and minimum length values in the products_pre_insert() function definition, but in reality one would probably want to create a table to store salt values as there should be a different salt value for each table that uses Hashids, and also salts should not be re-used between test environments and production.
Footnotes
1 I’m not saying it would necessarily affect SEO today, but SEO tends to trickle down from what Google et al consider to be human-friendly, which I don’t think excessively long machine-readable IDs are. I certainly think URLs that scroll way past the address bar with seemingly-random gibberish discourage people who share URLs via address bar copy and paste.
2 Although hash is in the name, the project makes clear it’s not a true cryptographic hash function (and thus not secure). But for my purposes, it’s exactly what I needed to discourage casual scraping while maintaining a certain level of user-friendliness that a very secure solution (UUIDs, real crypto hash functions) wouldn’t allow.
3 There are other Postgres extensions that add support for other languages, like PL/Python, but PL/V8 is a “trusted” Postgres language, while PL/Python is “untrusted.” Trusted languages are safer as they come with certain protections on what actions they can perform - untrusted languages can do anything that the database administrator can do! This is probably why AWS RDS supports PL/V8 but doesn’t support PL/Python.
Generating YouTube-like IDs in Postgres using PL/V8 and Hashids的更多相关文章
- A Deep Dive into PL/v8
Back in August, Compose.io announced the addition of JavaScript as an internal language for all new ...
- PostgreSQL-PL/pgSQL
参考: https://wiki.postgresql.org/wiki/9.1%E7%AC%AC%E4%B8%89%E5%8D%81%E4%B9%9D%E7%AB%A0 摘记: PL/pgSQL是 ...
- 通过ipv6访问 g o o g l e
Google.Youtube.Facebook等均支持IPv6访问,IPv4网络的用户大部分都无法访问,比如Gmail,Google Docs等等各种相关服务.而该类网站大部分均已接入IPv6网络,因 ...
- Consolidated Seed Table Upgrade Patch(Patch 17204589)
$ adop phase=apply patches= hotpatch=yes abandon=no Enter the APPS password: Enter the SYSTEM passwo ...
- shell 脚本阅读之二——ltp工具下的runltp
#!/bin/sh ################################################################################ ## ## ## ...
- <转>年终盘点!2017年超有价值的Golang文章
马上就要进入2018年了,作为年终的盘点,本文列出了一些2017年的关于Go编程的一些文章,并加上简短的介绍. 文章排名不分先后, 文章也不一定完全按照日期来排列.我按照文章的大致内容分了类,便于查找 ...
- kong服务网关API
kong服务网关API pingforever关注 0.1762017.05.23 11:16:08字数 834阅读 7,367 kong简介 Kong 是在客户端和(微)服务间转发API通信的API ...
- 【QQ技术】群文件报毒怎样下载?~ 变相绕过QQ复杂检验过程
刚才又人问我,要是群文件被鉴定为病毒那怎么下载? 我简单说一下吧: 其实qq客户端过滤比较严的,而web段却还是老一套,很多人说出现这个情况其实是腾讯已经把他库里面的文件删了,其实不然 如果源删了,那 ...
- (03)odoo模型/记录集/公用操作
-----------------更新时间11:17 2016-09-18 星期日11:00 2016-03-13 星期日09:10 2016-03-03 星期四11:46 2016-02-25 星期 ...
随机推荐
- 云服务器、VPS、虚拟主机三者之间的区别?
更多精彩内容欢迎访问我的个人博客皮皮家园:https://www.zh66.club期待您的光临哦!我是皮皮猪,感谢各位光临,能为您排忧解难小站深感荣幸!祝您生活愉快! 什么是虚拟主机? 同云主机不一 ...
- 《Interest Rate Risk Modeling》阅读笔记——第三章:拟合期限结构
目录 第三章:拟合期限结构 思维导图 扩展 第三章:拟合期限结构 思维导图 扩展 NS 模型的变种
- sql server删除重复记录只保留一条
今天遇到一个历史导入数据重复的问题,于是要删除重复的记录,一开始想用子查询的方式找到要删除记录的id删除,后来发现DELETE语句可以直接用外连接,这样更加简单,效率也更高. delete sys_p ...
- jquery validate 动态生成的多个同名input的验证
我的应用场景是,添加和修改入库单的明细,明细是以表格的形式呈现,可以动态添加商品,用jquery.validate插件做数据验证. 由于jquery.validate插件验证同名的input时只验证第 ...
- ServiceStack.Redis简单封装
首先创建RedisConfig配置类 #region 单例模式 //定义单例实体 private static RedisConfig _redisConfig = null; /// <sum ...
- C# 通过反射获取winform上的控件
比如获取Button按钮: System.Reflection.FieldInfo[] fieldInfo = form.GetType().GetFields(System.Reflection.B ...
- 如何写出优雅的 Golang 代码
原文: https://draveness.me/golang-101.html Go 语言是一门简单.易学的编程语言,对于有编程背景的工程师来说,学习 Go 语言并写出能够运行的代码并不是一件困难的 ...
- VBA笔记
1.VBA数据类型 数据类型 存储空间大小 范围 Byte 1个字节 0-255 Boolean 2个字节 true或false Integer 2个字节 -32768-32767 Long 4个字节 ...
- Linux从入门到放弃、零基础入门Linux(第二篇):在虚拟机vmware中安装linux(一)超详细手把手教你安装centos分步图解
一.Vmware vmware介绍:VMware,Inc. (Virtual Machine ware)是一个“虚拟PC”软件公司,提供服务器.桌面虚拟化的解决方案.其虚拟化平台的产品包括播放器:它能 ...
- Semaphore的简介及应用场景
Semaphore是一个计数信号量,常用于限制可以访问某些资源(物理或逻辑的)线程数目. 常用函数: 信号量的构造函数 非公平: public Semaphore(int permits);//per ...