index.rst (6116B)
1 ==== 2 AWSY 3 ==== 4 5 Are We Slim Yet project (commonly known as AWSY) tracks memory usage across builds. 6 7 On treeherder, the AWSY builds are listed in subgroups of `SY`. 8 9 AWSY tests consist of three types: TP5*, TP6, and Base Memory Usage. 10 11 *\*TP5 tests are out of date and no longer maintained. These tests are scheduled to be removed:* https://bugzilla.mozilla.org/show_bug.cgi?id=1712406 12 13 {documentation} 14 15 Running AWSY Locally 16 ********************* 17 18 Running tests locally is most likely only useful for debugging what is going on in a test, 19 as the test output is only reported as raw JSON. The CLI is documented via: 20 21 .. code-block:: bash 22 23 ./mach awsy-test --help 24 25 *Currently all tests will download TP5 even if it is not used, see:* https://bugzilla.mozilla.org/show_bug.cgi?id=1683920 26 27 TP5 tests 28 ========= 29 30 .. code-block:: bash 31 32 ./mach awsy-test --tp5 33 34 TP6 tests 35 ========= 36 37 .. code-block:: bash 38 39 ./mach awsy-test 40 41 Base Memory Usage tests 42 ======================== 43 44 .. code-block:: bash 45 46 ./mach awsy-test --base 47 48 Running AWSY on Try 49 ******************** 50 51 AWSY runs can be generated through the mach try fuzzy finder: 52 53 .. code-block:: bash 54 55 ./mach try fuzzy 56 57 A query for "awsy" will return all AWSY tests. The default test is TP5, TP6 and Base test names will contain `tp6` and `base`, respectively. 58 59 The following documents all tests we currently run for AWSY. 60 *The following content was migrated from* https://wiki.mozilla.org/AWSY/Tests *and will be updated to clarify TP5/TP6 tests vs Base tests:* https://bugzilla.mozilla.org/show_bug.cgi?id=1714600 61 62 63 TP5/TP6 Tests 64 ************** 65 66 The following tests exist for both TP5 and TP6. Running `./mach awsy-test` by default will run TP6 tests*. 67 To run TP5 tests, add the `--tp5` flag: `./mach awsy-test --tp5` 68 69 *\*TP5 tests are out of date and no longer maintained. These tests are scheduled to be removed:* https://bugzilla.mozilla.org/show_bug.cgi?id=1712406 70 71 72 Explicit Memory 73 ================ 74 75 * This is memory explicitly reported by a memory reporter. It includes all the memory allocated via explicit calls to heap allocation functions (such as malloc and new), and some (only that covered by memory reporters) of the memory allocated via explicit calls to non-heap allocation functions (such as mmap and VirtualAlloc). 76 77 **Possible regression causes** 78 79 * A regression in this usually means a new feature is using or retaining more memory and should be looked at. These are easier to diagnose because we can compare memory reports. 80 81 See the `about:memory` `mdn page <https://developer.mozilla.org/docs/Mozilla/Performance/about:memory#Explicit_Allocations>`__ for more details. 82 83 84 Heap Unclassified 85 ================== 86 87 *to do: add test definition* 88 89 90 Images 91 ======= 92 93 * This is a subset of the "explicit" measurement that focuses on memory used to render images. 94 95 **Possible regression causes** 96 97 * A regression in this can indicate leaks or poor memory usage in the image subsystem. In the past this was a persistent problem. 98 99 100 JS 101 ==== 102 103 *to do: add test definition* 104 105 106 Resident Memory 107 ================ 108 109 * This is a higher level measurement provided by the operating system. We sum the "resident" memory (`RSS <https://en.wikipedia.org/wiki/Resident_set_size>`_) with the `resident-unique <https://en.wikipedia.org/wiki/Unique_set_size>`_ memory of the content processes. It's pretty noisy and large so it's not very useful in detecting smaller regressions. 110 111 **Possible regression causes** 112 113 * Regressions in this often track regressions in explicit and heap unclassified. If we see a regression in resident, but not in other reports this can indicate we are leaking untracked memory (perhaps through shared memory, graphics allocations, file handles, etc). 114 115 116 Base Content Tests 117 ******************* 118 119 * An updated test focused on supporting Fission. This measures the base overhead of an empty content process. It tracks resident unique, heap unclassified, JS, and explicit memory metrics as well as storing full memory reports as artifacts. The median value for each metric is used from across all content processes. It has much lower thresholds for alerting and is recorded in `Perfherder <https://wiki.mozilla.org/EngineeringProductivity/Projects/Perfherder>`_. 120 121 122 Base Content Explicit 123 ====================== 124 125 **Possible regression causes** 126 127 A change has caused more JavaScript to load at startup or into blank pages. 128 129 * **Common solution**: lazily load any new modules you rely on 130 * **Common solution**: Split your code out to only load what is minimally needed initially. You modified the JS engine and it's using more memory 131 * **Common solution**: Attempt to reduce your object size for the common case, these tend to add up! You implemented a new feature in JavaScript 132 * **Common solution**: Write the majority (or all of it) in compiled code (C++/Rust). This will reduce overhead and generally improve performance. 133 134 135 Base Content Heap Unclassified 136 =============================== 137 138 * The "heap-unclassified" value represents heap-allocated memory that is not measured by any memory reporter. This is typically 10-20% of "explicit". 139 140 141 **Possible regression causes** 142 143 * A regression in this can indicate that we're leaking memory or that additional memory reporters should be added. 144 * An improvement can indicate that leaks have been fixed or that we added new memory reporters. 145 146 See the `about:memory` `mdn page <https://developer.mozilla.org/docs/Mozilla/Performance/about:memory#Explicit_Allocations>`__ for more details. 147 148 149 Base Content JS 150 ================ 151 152 * This is the "js-main-runtime/" value in `about:memory` which is all the memory attributed to the javascript engine. 153 154 **Possible regression causes** 155 156 * A regression in this number can indicate leaks in the JS engine, optimizations that take performance into consideration at the expense of more memory, or problems with the garbage collector. 157 158 159 Base Content Resident Unique Memory 160 ==================================== 161 162 *to do: add test definition* 163 164 165 Other references 166 ----------------- 167 168 `Are We Slim Yet MDN web docs <https://developer.mozilla.org/en-US/docs/Mozilla/Performance/AWSY>`_