Squash: A squarish bug spray
An open-source project from Square
Squash is a collection of tools that help engineers find and kill bugs in their code by automatically collecting, collating and analyzing run time exceptions. Squash consists of the following components:
- Client libraries: Client libraries for different systems (Ruby, Ruby on Rails, Cocoa with Objective-C, etc.) catch and record errors when they occur, then send them to the API endpoint. Client libraries can be found under the SquareSquash organization.
- Front-end: This website displays Bug information and helps the engineer find the root cause for a Bug, and fix it. It also lets engineers manage, assign, and comment on Bugs, as well as view statistics about the Bug.
- API endpoints: These routes (part of the front-end app) receive exception notifications from the client libraries and process them.
This project is the front-end and the API.
Pull requests are more than welcome; please check out CONTRIBUTING.md for details.
How to Install
To get started, run the
setup.rb file in the project root. This script will
ask you a series of questions and generate a preliminary configuration for your
site install. When the script is completed, you can run
git status to see what
files it changed, and refine your configuration from there.
Once the script is complete, Squash should run for most typical development
environments. Simply run
rails s and visit the site in your Web browser. You
should be able to start using it immediately. You can also verify correctness by
Configuring and deploying the production instance is entirely up to you and your particular production environment.
Additional configuration options can be found in the following locations:
If you don't see what you're looking for in any of those files, you'll probably have to change the code to make it work. Don't be afraid -- the code is thoroughly documented and should (hopefully) be very accessible.
Squash requires the following:
- Ruby 1.9.2 or newer (JRuby with
- Multithreading support (see the next section)
- PostgreSQL 9.1 or newer
- The Bundler gem
- Git 1.7 or newer
Notes on some of the gem and library choices
Why do you specifically require PostgreSQL? Squash uses a lot of PostgreSQL-specific features to make efficient use of the database and maintain referential integrity, such as:
- foreign key constraints,
- triggered cached counters,
- check constraints, and
- semantic indexes for text-based search.
If PostgreSQL is out of the question, some of these features can be ported to
other RDBMSes; simply edit the
InitialSchema migration and update the SQL as
necessary. If portability is required, a lot of these features can be
reimplemented in the Rails layer (e.g., cached counters), at the risk of
degraded referential integrity. (If you are reimplementing the cached counters
in Ruby, be sure to modify the
config/initializers/active_record.rb file as
If you do successfully port Squash to another RDBMS, let me know. I'd be happy to take your changes.
Why do you bundle an edge version of Rails? The
3-2-stable branch of Ruby
on Rails includes some changes to Active Record that are required by Squash's
multithreaded concurrency model (see next question). In particular, that version
of Active Record includes crucial changes to the connection pool and connection
reopening logic. If you set the
background_runner option in the
concurrency.yml file to something other than
Multithread, the Gemfile will
automatically drop Rails back down to the latest release version.
Why don't you use my favorite backgrounding library? Squash was originally built for Square, which runs all its services on JRuby. Using threads is very efficient in JRuby, and avoids the overhead of having to deploy both a website and workers.
If you are running Squash in a non-thread-safe (or multithreading-unfriendly)
environment, you can use Sidekiq or Resque instead. If you want to use some
other backgrounding library, you can easily write your own adapter. All threaded
code is invoked using
concurrency.yml Configoro file to invoke the correct module under
lib/background_runner. The default is to use ,
which uses the module to execute the task in its own thread. You
can edit the YAML file and switch the background runner to Resque, or implement
BackgroundRunner module. See the documentation for
If you do this successfully and wish to save future Squash users the effort, feel free to turn your changes into a pull request.
Why aren't you using RedCarpet? As mentioned above, Squash was originally built to run under JRuby. RedCarpet has compiled C extensions; Kramdown is pure Ruby.
Why do you require a recent version of Git? Squash uses the
git clone --mirror command to create local mirrors of client projects' Git
Why don't you have integration tests or acceptance tests? To be 100% honest, lack of familiarity on these things in Ruby/Rails. Happy for any help people want to extend towards this goal.
Why are you using Erector? I like Erector.
Comprehensive documentation is written in YARD- and Markdown-formatted comments
throughout the source. To view this documentation as an HTML site, run
CoffeeScript libraries are documented using the YARD format as well, but YARD
does not as yet recognize them as documentable files. A
.codoopts file is
included in case you wish to use Codo to
generate the CoffeeScript docs, but as of now Codo does not recognize the ERb
files, and does not use the full set of Markdown syntax features used in the
This is a pretty typical Rails website, save for the views, which are written using Erector. The views forgo the traditional Rails concepts of partials and templates in favor of analogous OOP concepts more familiar to software developers: methods and inheritance. All views inherit from an abstract Erector widget which provides layout; and all views have their content split into multiple private methods.
In addition to the usual helpers (in
app/helpers), there are view mixins under
app/views/additions that simplify view coding.
Embedded code snippets are all rendered using the
are documented but the documentation is not recognized by YARD and so is not
included in this documentation set.
site area are in
behavior to a specific page is in a
.jsfile named the same as, and placed alongside, the
.html.rbview file. For example, if
app/views/projects/new.html.rbneeded a bit of JS glue code, it would be placed in
app/views/projects/new.js. This code is placed in a
<SCRIPT>tag at the end of the view by the method.
CSS files are similarly organized:
- Third-party CSS files are in
vendor/assets/stylesheetsand loaded in the
- CSS styles or helpers global to the entire website are in
lib/assets/stylesheetsand also loaded in
- CSS styles specific to a single page or a related group of pages are placed in
app/assets/stylesheetsand also loaded in
<BODY>tag is given a class name equal to the controller name, and an ID equal to the controller and action name separated with a dash. For instance, the
projects/newaction's body would be
<body class=projects id=projects-new>.
For information about requests and responses, see.
Models make extensive use of advanced PostgreSQL features for efficiency and
convenience. Cached counters are updated using triggers and rules, foreign key
constraints and hooks are enforced at the database level, and validations are
backed up by corresponding
CHECK triggers. This helps ensure referential and
data integrity even in situations where Rails fails, or outside of the Rails
stack. See the various migrations to learn more about the triggers, rules, and
constraints being used.
Observers are used for more high-level triggers, such as creating
app/models/observers for more.
Models also use the HasMetadataColumn gem to reduce their width and incorporate
schemaless data. Most models have a JSON-formatted
metadata column to which
new information can be added or removed without having to create new migrations.
Various Rake tasks are available under
lib/tasks. These include tasks for
pruning and maintaining the database, development tasks, and configuring the
Workers are found in the
lib/workers directory. Along with OccurrencesWorker,
which stores and categorizes Occurrences, there are other workers for managing
Deploys and other minor tasks. These workers are run asynchronously using
Notification mails are sent by the
NotificationMailer conditionally delivers emails. An email will only be
delivered if all of the following conditions are met:
- The Project has a mailing list email configured (for the critical-bugs and all-bugs mailing lists),
- the Environment's
- the User has enabled receipt of that category of emails (typically specified in the Membership).
Authentication and Authorization
Authentication is done using either password verification or LDAP; see
app/models/additions for more information.
There are four permissions levels that a User can have, specific to an individual Project:
Non-members do not have arecord with a Project. They can view Bugs and Occurrences, view the Project's API key, view the list of other Project members, watch Bugs, and comment on Bugs.
Members can do everything non-members can do, and can also assign Bugs, be assigned Bugs, and modify/delete Bugs.
Administrators can do everything members can do, and can also modify Project and Environment settings, regenerate the API key, promote and demote members to administrator status, and modify/delete others' Comments.
Owners (each Project has only one) can do everything administrators can do, and can also delete the Project and reassign ownership.
Recording and Categorizing Occurrences
The client library identifiers used throughout the website determine how a Bug reported from that library is presented in the view. Theconstant maps a client library identifier to the relevant summary fields to display in the list view.
Each occurrence is transmitted with the name of the client library; the
client field. The 's
client field is
set from the first Occurrence's; in general, one should expect that all
Occurrences of a Bug share the same client value.
Regardless of the Occurrence's client value, all fields for which there is data are displayed in the Occurrence view.
If a client library is updated to add new information fields, all that is needed
is to update the
has_metadata hash in Occurrence with the new fields, and to
occurrences#show view as necessary. (The
INDEX_FIELDS hash can
also be updated as appropriate.)
If a new client library is added, in addition to doing the above for any new
fields unique to the new client, the
INDEX_FIELDS hash will need to be
expanded to include the new client.
For information about the background worker that converts incoming exception information into Occurrence and Bug records (including doing "best guess" commit blaming, determining which Occurrences share the same root Bug, etc.), see the documentation for themodule. See also Static Analysis below.
Deploys and Releases
Squash can handle both deployed projects (hosted projects for which typically only one version is live at a time) and released projects (distributed apps for which many versions may exist "in the wild").
If you are developing a released project, you must associate yourobjects with a unique version identifier (such as a build number). You must also send the correct build identifier with every occurrence report. See the client library and API controller documentation for more information.
Once this is done, Squash's behavior is changed slightly: Bugs that are fixed in an older release, but recur, are not reopened. Bugs that are fixed in an older release but then recur in newer releases are treated as completely separate Bugs.
This app exposes an API endpoint for notifying Squash of new releases/deploys of your Project. You should notify Squash whenever you deploy your Project (or release a new version) so that it can properly manage Bugs (automatically mark fixed Bugs as deployed, associate new Bugs with the new Deploy, etc.).
Most client libraries include an easy means to add this feature to your deploy or release process.
When an Occurrence is grouped into a Bug, its message is stripped of any non-relevant or situational information. (Note that since the message is not used as a dimension in grouping Occurrences, two Occurrences of the same Bug could have completely different messages. A Bug gets its message from its first Occurrence.) This is done by subclasses of.
For most messages, this is done using simple regex subsitution. Squash can also
normalize an exception's message using a list of known message templates. Such
templates are stored in the
data/message_templates.yml file. The file contains
a hash mapping exception class names to an array of arrays. Each inner array
contains two elements: the regexp to match the exception message on, and the
replacement message should it match.
These templates are evaluated in the order they appear in the array. Scripts
that regenerate the templates for MySQL and PostgreSQL error strings can be
found under the
Version Control Systems
As of now, Squash is only compatible with projects that are version-controlled using Git. However, in the interest of remaining as VCS-agnostic as possible, commit identifiers are never called "SHA1s," but instead "IDs."
Unsymbolicated, Obfuscated, or Minified Stack Traces
Client libraries of compiled, obfuscated, or minified applications will need to convert their stack traces in order for them to be of use for static analysis. Because release builds are typically not distributed with embedded decompilation information, a mapping or table must be given to Squash, so that when Squash receives an unconverted stack trace, it can perform lookups and convert it to a more usable format.
Currently, the supported conversions are:
- symbolication of iOS exceptions (see ),
- deobfuscation of Java exceptions (see ),
Client libraries are responsible for delivering the raw stack trace data to Squash when an exception occurs, and for delivering lookup tables to Squash upon each new release.has endpoints for these purposes in various languages.
Unconverted stack traces are stored in a particular format; seefor more information. When a new exception with a stack trace in this format is received, Squash immediately attempts to convert it using an existing matching lookup table. If no such table is found, the stack trace is left unconverted. It can still be viewed on the front end. When, later, a new lookup table is added, Squash automatically finds and converts any matching stack traces.
Because every new Occurrence must be assigned a Bug (including unconverted
Occurrences), it is possible that the "blamed"
line fields of the
Bug could themselves be unconverted. The class has provisions to support
this; see in particular
All models, controllers, and library files are unit-tested with RSpec specs
spec directory. Run unit tests with the
rspec spec command. Views
written. Almost all unit tests use factories rather than mocks, putting them
somewhat closer to integration tests.
In general, the test environment is identical to the development/production
environment, save for the usual Rails allowances, and the
config/initializers/active_record_observer_hooks.rb file. This file adds
after-commit hooks to observers (identical to those available to models).
However, specs are transactionalized, meaning that these hooks wouldn't run
until after the spec is completed, resulting in multiple spec failures.
To remedy this, the file introspects on the environment, and instead links the
observer hooks to the
after_save model hooks in the test environment. This
differentiates the test and live environments, but allows specs to pass.
If you wish to bring your test environment closer to production, you can set
use_transactional_fixtures RSpec setting to false. The hooks file will
automatically detect the change and use the correct
Currently, all specs pass with and without transactionalized fixtures.
Squash supports integration with a handful of popular other developer tools.
When Squash is integrated with PagerDuty, it will begin sending exceptions to PagerDuty once the Occurrence count for a given Bug exceeds the Project's critical threshold. Thereafter, every Occurrence is sent to PagerDuty as a trigger. The triggers are grouped into alerts by Bug, so users will only be paged once per new error.
In addition, marking a Bug fixed will also automatically resolve its associated PagerDuty alert (if any), and assigning a Bug or marking it irrelevant automatically acknowledges any associated PagerDuty alert.
In order to integrate Squash with PagerDuty, you must
- generate an API key that Squash can use for PagerDuty access (you must be a PagerDuty admin to do this),
- update the
config/environments/common/pagerduty.ymlfile to enable PagerDuty integration and provide the API key,
- configure Generic API services for each escalation policy you wish to use, and
- set the PagerDuty-related options in your Project's configuration page.
When Squash is integrated with JIRA, users have the option of associating a JIRA issue with any Bug. The Management tab will then display a link allowing the user to quickly view the associated issue.
In addition, the user can have Squash watch the JIRA issue for status updates. Once the JIRA issue is resolved (or is changed to someother specified status), the Bug can be automatically marked as fixed. By associating multiple Bugs with one JIRA issue in such a manner, a user can quickly fix a large number of Bugs by closing just one JIRA issue.
Finally, a "Create JIRA Issue" button is provided on the Management tab allowing users to easily create new JIRA issues for any Bug.
In order to integrate Squash with JIRA, you must
- configure the
config/environments/common/jira.ymlfile with your JIRA authentication credentials and installation location, and
- add a cron job or other periodic task that runs
JIRA integration works out-of-the-box if you use username/password
authentication. OAuth authentication requires a few more steps. Because OAuth
authentication requires user actions, you will need to obtain an access token
and secret which the JIRA client can use without needing to prompt the user. You
can do this by running the
rails runner script/jira_oauth.rb
Follow the instructions to set up your JIRA integration.