ruby

Practical Garbage Collection Tuning in Ruby

Guest Author

Guest Author on

Practical Garbage Collection Tuning in Ruby

We have discovered that the below post is based on an article by Nate Berkopec from 2017 called 'Understanding Ruby GC through GC.stat'. It appears that parts of this article were plagiarised, something we were unaware of until the original author mentioned it. We run all of our articles through a plagiarism tool before publishing, but it didn't pick this up. We give huge apologies to Nate and our readers for this inadvertent error.

It is vital that you understand how garbage collection works in Ruby to stay in complete control of your app's performance.

In this post, we will dive into how to implement and customize garbage collection in Ruby.

Let's get going!

The Ruby Garbage Collector Module

The Ruby Garbage Collector module is an interface to Ruby's mark and sweep garbage collection mechanism.

While it runs automatically in the background when needed, the GC module lets you call the GC manually whenever required and gain insights into how garbage collection cycles are running. The module provides some parameters which you can alter to moderate performance.

Some of the most commonly used methods of this module are:

  • start/garbage_collect: This method initiates the garbage collection cycle manually.
  • enable/disable: These methods enable or disable the automatic garbage collection cycles. They return a boolean value indicating if the operation was successful.
  • stat: This method provides a list of keys and values that describe the performance of the GC module. We will take a look at these metrics in detail in the next section.

Understanding Ruby Garbage Collector Parameters

To understand how Ruby’s GC works internally, let’s look at the GC module's metrics. Run the following command on a freshly booted irb:

ruby
puts GC.stat

You will notice that a bunch of numbers pop up on your screen, looking something like this:

ruby
{ :count=>12, :heap_allocated_pages=>49, :heap_sorted_length=>49, :heap_allocatable_pages=>0, :heap_available_slots=>19975, :heap_live_slots=>19099, :heap_free_slots=>876, :heap_final_slots=>0, :heap_marked_slots=>16659, :heap_eden_pages=>49, :heap_tomb_pages=>0, :total_allocated_pages=>49, :total_freed_pages=>0, :total_allocated_objects=>66358, :total_freed_objects=>47259, :malloc_increase_bytes=>16216, :malloc_increase_bytes_limit=>16777216, :minor_gc_count=>10, :major_gc_count=>2, :remembered_wb_unprotected_objects=>191, :remembered_wb_unprotected_objects_limit=>312, :old_objects=>16024, :old_objects_limit=>23556, :oldmalloc_increase_bytes=>158824, :oldmalloc_increase_bytes_limit=>16777216 }

This holds all the information about how garbage collection has been happening in the runtime. Let's examine each of these numbers in detail.

Counts in Ruby Garbage Collector

We'll begin by describing these keys:

ruby
{ :count=>12, #… :minor_gc_count=>10, :major_gc_count=>2, }

These are GC counts, and they convey pretty straightforward information. minor_gc_count and major_gc_count are the counts of each type of garbage collection run.

There are two types of garbage collections in Ruby.

Minor GC refers to a garbage collection attempt that tries to garbage collect only those objects that are new, i.e., they have survived three or fewer garbage collection cycles.

On the other hand, major GC is a garbage collection attempt that tries to garbage collect all objects, even those that have survived more than three garbage collection cycles. count is the sum of minor_gc_count and major_gc_count.

Tracking the GC count can be helpful for a few reasons. You can figure out if a particular job or process always triggers GCs and the number of times it triggers them. It might not be 100% accurate in cases like multithreaded applications, but it is a good starting point to figure out where your memory is bleeding.

Heap Numbers: Slots and Pages

Next, let’s talk about these keys, also known as heap numbers:

ruby
{ # page numbers :heap_allocated_pages=>49, :heap_sorted_length=>49, :heap_allocatable_pages=>0, # slots :heap_available_slots=>19975, :heap_live_slots=>19099, :heap_free_slots=>876, :heap_final_slots=>0, :heap_marked_slots=>16659, # Eden and Tomb pages :heap_eden_pages=>49, :heap_tomb_pages=>0, }

The heap that we are talking about here is a C data structure. It contains references to all the currently live Ruby objects. A heap page is composed of memory slots, and each slot includes information on only one live Ruby object:

  • heap_allocated_pages is the number of currently allocated heap pages. These pages can be completely empty, completely filled, or partly filled.
  • heap_sorted_length is the actual size that the heap has occupied in memory and is different from heap_allocated_pages, as the length is the length of the heap pages put together, not their count. If you initially allocated 10 pages and then freed one from the middle of the set, your heap_allocated_pages would be 9, but the heap_sorted_length would still be 10.
  • Finally, heap_allocatable_pages is the count of heaps that Ruby currently owns that can be used when needed.

Now, coming to the slots:

  • heap_available_slots is the total number of available slots in the heap pages.
  • heap_live_slots is the number of live objects in the memory.
  • heap_free_slots are slots in the allocated heap pages that are empty.
  • heap_final_slots is the count of slots whose objects have finalizers attached to them. Finalizers are Procs that run when an object is freed, similarly to destructors in OOPS.
  • heap_marked_slots is the count of old objects (i.e., the objects that have been around for more than 3 GC cycles) and write-barrier unprotected objects.

Then we have tomb_pages and eden_pages.

tomb_pages is the count of pages that contain no live objects. These pages are eventually released back to the operating system by Ruby.

On the other hand, eden_pages is the count of those pages that contain at least one live object, so they can't be released back to the operating system.

Consider monitoring the metric heap_free_slots if you face memory bloat issues in your application.

A high number of free slots (more than 250,000) usually indicates that you have a handful of controller actions that allocate many objects at once and then free them. This can permanently bloat the size of your running Ruby process.

Cumulative Numbers

ruby
{ :total_allocated_pages=>49, :total_freed_pages=>0, :total_allocated_objects=>66358, :total_freed_objects=>47259, }

These numbers are cumulative or additive in nature for the entire life of the process. They are never reset by the GC and can't technically go down. All four of these numbers are self-explanatory.

Garbage Collection Thresholds

To understand these numbers, you first need to understand when GC is triggered:

ruby
{ :malloc_increase_bytes=>16216, :malloc_increase_bytes_limit=>16777216, :remembered_wb_unprotected_objects=>191, :remembered_wb_unprotected_objects_limit=>312, :old_objects=>16024, :old_objects_limit=>23556, :oldmalloc_increase_bytes=>158824, :oldmalloc_increase_bytes_limit=>16777216 }

Contrary to a common assumption that GC runs happen at fixed intervals, GC runs are triggered when Ruby starts running out of memory space. Minor GC happens when Ruby runs out of free_slots.

If Ruby is still low on free_slots after a minor GC run — or the threshold of oldmalloc, malloc, old object count, or shady/write-barrier-unprotected count is exceeded — a major GC run is triggered. The above part of gc.stat shows the values of these thresholds.

malloc_increase_bytes refers to the amount of memory allocated outside of the heap we talked about so far. When the size of an object exceeds the standard size of a memory slot — say, 40 bytes — Ruby mallocs some space somewhere else just for that object. When the total extra allocated space exceeds malloc_increase_bytes_limit, a major GC is triggered.

oldmalloc_increase_bytes is a similar threshold for old objects. old_objects is a count of object slots marked as old. When the number exceeds the old_objects_limit, major GC is triggered.

remembered_wb_unprotected_objects is the total count of objects that are not protected by the write-barrier and are a part of the remembered set.

The write-barrier is an interface between Ruby runtime and its objects, which allows the interpreter to track references to and from the object as soon as they are created.

C-extensions can make new references to objects without using the write-barrier, in which case the objects are marked shady or write-barrier unprotected. The remembered set is simply a list of old objects with at least one reference to a new object.

Customize Ruby Garbage Collection Performance

Now that you understand how Ruby GC manages your application's memory, it is time to look at the options available to customize GC's behavior.

Here are environment variables that you can use to moderate the performance of Ruby GC and, in turn, improve the performance of your application:

bash
RUBY_GC_HEAP_INIT_SLOTS RUBY_GC_HEAP_FREE_SLOTS RUBY_GC_HEAP_GROWTH_FACTOR RUBY_GC_HEAP_GROWTH_MAX_SLOTS RUBY_GC_HEAP_OLDOBJECT_LIMIT_FACTOR and other variables

Let's talk about the important parameters here, one by one:

  • RUBY_GC_HEAP_INIT_SLOTS: defines the initial number of slots on the Ruby heap and is set to 10000 by default. You might want to change this parameter if you're sure that your app will initially allocate most of its objects.
  • RUBY_GC_HEAP_FREE_SLOTS: controls the minimum number of free slots that must be available right after a GC cycle. Its default value is 4096. This value is only used once at runtime during the first heap growth.
  • RUBY_GC_HEAP_GROWTH_FACTOR: the factor by which available heaps to the Ruby interpreter grow. Its default value is 1.8. Changing this makes little sense, as Ruby is already aggressive at heap growth. It will not make much difference if you try decreasing it since heaps are allocated on-demand in modern interpreters.
  • RUBY_GC_HEAP_GROWTH_MAX_SLOTS: the maximum number of slots Ruby can add to the heap space at once. The default value is 0, which stands for no limit on the number. If your app needs to allocate millions of objects in its lifetime, you might want to put a cap on this parameter. However, it will have a pretty low effect on the GC time of your app.
  • RUBY_GC_HEAP_OLDOBJECT_LIMIT_FACTOR forces the interpreter to carry out a major GC cycle when the total number of old objects in memory = more than this number x number of old objects in memory after the last GC cycle. You might want to increase this number if you think many of your objects will become unused after they enter the old generation. However, this is rarely needed.
  • RUBY_GC_MALLOC_LIMIT is the minimum limit of malloc calls for the new generation. Its default value is 16 MB. RUBY_GC_MALLOC_LIMIT_MAX is the maximum limit for the same malloc calls. Its default value is 32 MB. You might want to increase these two limits if your application uses higher than average memory. However, be careful not to raise these too much. Otherwise, they might lead to higher peak memory consumption. Always increase these limits incrementally — say, by 4 or 8 MB.
  • RUBY_GC_MALLOC_LIMIT_GROWTH_FACTOR is the growth factor of the malloc limit for the new generation. Its default value is 1.4. You should consider increasing this number if your app allocates memory in chunks rather than whole at once.
  • Similarly, RUBY_GC_OLDMALLOC_LIMIT and RUBY_GC_OLDMALLOC_LIMIT_MAX are the minimum and maximum malloc limits for the old generation. The default values of these parameters are 16 MB and 128 MB.
  • RUBY_GC_OLDMALLOC_LIMIT_GROWTH_FACTOR is the growth factor of this limit. Its default value is 1.2. You can consider changing these with the new generation limits for maximum effect.

Fine-tuning Garbage Collection in Ruby

We discussed some common and simple ways to customize the GC module to help you improve your application's overall performance. However, these tweaks might not work in all cases. You need to figure out the memory usage pattern of your app before deciding what to customize.

On the flip side, you can consider running an automated test that finds the best values of these parameters for you. Tools like TuneMyGC are pretty straightforward when figuring out the best set of values for your environment variables.

Definitely look at GC parameters if your application is behaving weirdly. A small change here can go a long way to bring down your app's memory consumption and prevent memory bloats.

I hope this article has given you a good idea of what to look out for when customizing your Ruby Garbage Collection module. For more of an introduction, check out the Introduction to Garbage Collection Part I and Part II.

Have fun coding!

Become our next author!

Find out more

AppSignal monitors your apps

AppSignal provides insights for Ruby, Rails, Elixir, Phoenix, Node.js, Express and many other frameworks and libraries. We are located in beautiful Amsterdam. We love stroopwafels. If you do too, let us know. We might send you some!

Discover AppSignal
AppSignal monitors your apps