test cache page allocation failure

[goatshriek]

Stumpless uses a simple slab cache implementation in src/cache.c to consolidate memory allocations for common structures like entries. Test coverage of this cache is fairly complete, but still has some gaps.

One such gap is the failure of a new page allocation when the cache needs one. You will test for this case by causing a memory allocation failure after the cache has been set up and needs to add a new page.

General Approach

There are a few details left out of the following approach, for you to fill in as you encounter them. If you find you need help, please ask here or on the project gitter and someone can help you get past the stumbling block.

First, familiarize yourself with the cache_alloc implementation in src/cache.c. This function looks for a free slot in the cache, and if one is not found then it allocates a new page to make more space. You don't need to fully understand this function, but it will help to see what logic causes the add_page call to happen, which is what you will cause to fail.

Next, devise a test that creates a single structure in a cache (entry structures are one such type) to set up the cache, set the memory allocation function to one that will fail, and then create new structures until the cache fills, attempts to allocate a new page, and fails. For an example of a test that creates several new structures look at TEST( BufferTargetOpenTest, Open100Targets ) in test/function/target/buffer.cpp, and TEST_F( EntryTest, AddNewParamMallocFailure ) in test/function/entry.cpp for an example of a memory allocation failure test. If testing entry creation, put your new test in test/function/entry.cpp.

Make sure that you have full coverage on add_page in src/cache.c after this test. See the test documentation for details on checking coverage.

[rrrrrrmb]

Hi, I'd like to work on this. I haven't really done much in C but would like to try. Based on my understanding the line of interest to us is

new_page = alloc_mem( c->page_size ); in add_page in src/cache.c

This new page allocation should fail in our case however I'm not sure how to set this up with MALLOC_FAIL_ON_SIZE since I don't know how to get the page_size value. Is there another way to do this ? Can you give pointers to solve this ?

Thanks !

[goatshriek]

It is a tricky one to understand and test for, which is why there isn't a test now!

The way the test described in the issue works is to first set things up normally, allowing memory to be allocated as needed. After this, you will set the memory to one that fails in any scenario (no need to worry about the size, just use MALLOC_FAIL), and then create new objects that are backed by a cache. entry structures are a good candidate here, and creating a new one with stumpless_new_entry_str with a NULL message should work while the cache has space left, even if the malloc would fail. In the test I would create a new entry this way normally, then set malloc to fail, and then loop on entry creation until it fails. I believe this will exercise the code path in question.

[goatshriek]

Resolved by #464.