Robj/workflow parallelism (#656)

* remove experimental status on shared memory support
* new test infra
  * increase parallelism and reduce waste
  * clean up list of tests to run
  * use my version of space maximizer to workaround bugs in upstream
* fix deadlock in shard_log_write
* make clockcache_get_free_page always succeed
* fix log_test --perf forced failure exit code

---------

Signed-off-by: Rob Johnson <rob@robjohnson.io>

Author: rtjohnso

.github/workflows/run-tests.yml

@@ -19,66 +19,52 @@ jobs:
   run-test-matrix:
     strategy:
       matrix:
-        compiler_mode: [gcc, clang, gcc-asan, clang-msan]
-        build_mode: [release, debug]
-        target: [all] # Just compile most configs
-        include_slow_tests: ['true']
-        run_nightly_tests: ['false']
+        compiler_mode: [clang-msan, gcc-asan, clang, gcc]
+        build_mode: [debug, release]
+        target: [run-tests]
+        tests_function: [cache_tests_1, cache_tests_2, cache_tests_3, functionality_tests, parallel_perf_test_1, parallel_perf_test_2, parallel_perf_test_3, parallel_perf_test_4, perf_tests_1, perf_tests_2, splinter_misc_tests_1, splinter_misc_tests_2, large_insert_stress_tests_1, large_insert_stress_tests_2, large_insert_stress_tests_3, misc_tests, filter_tests, unit_tests]
         exclude:
-          # Don't do a compile job on these, since we will do run-tests on them below
           - compiler_mode: gcc-asan
-            build_mode: release
-            target: all
+            build_mode: debug
+            target: run-tests
           - compiler_mode: clang-msan
-            build_mode: release
-            target: all
-          - compiler_mode: clang
             build_mode: debug
-            target: all
-          - compiler_mode: gcc
-            build_mode: release
-            target: all
+            target: run-tests
+          # The following tests are excluded because they take too long to run with sanitizers
+          - build_mode: debug
+            tests_function: large_insert_stress_tests_2
+          - build_mode: debug
+            tests_function: large_insert_stress_tests_3
+          - compiler_mode: gcc-asan
+            tests_function: large_insert_stress_tests_2
+          - compiler_mode: gcc-asan
+            tests_function: large_insert_stress_tests_3
+          - compiler_mode: clang-msan
+            tests_function: large_insert_stress_tests_2
+          - compiler_mode: clang-msan
+            tests_function: large_insert_stress_tests_3
         include:
-          # Compile and run tests on these configs
           - compiler_mode: gcc-asan
-            build_mode: release
-            target: run-tests
-            include_slow_tests: 'true'
-            run_nightly_tests: 'false'
+            build_mode: debug
+            target: all
           - compiler_mode: clang-msan
-            build_mode: release
-            target: run-tests
-            include_slow_tests: 'true'
-            run_nightly_tests: 'false'
-          - compiler_mode: clang
             build_mode: debug
-            target: run-tests
-            include_slow_tests: 'true'
-            run_nightly_tests: 'false'
-          - compiler_mode: gcc
-            build_mode: release
-            target: run-tests
-            include_slow_tests: 'false'
-            run_nightly_tests: 'true'
-    name: >
-          ${{ matrix.target == 'all' && 'Compile' || 'Run' }}
-          ${{ matrix.compiler_mode }} ${{ matrix.build_mode }} ${{ matrix.target }}
-          ${{ matrix.target == 'run-tests' && matrix.include_slow_tests == 'true' && '(slow tests)' || '' }}
-          ${{ matrix.target == 'run-tests' && matrix.run_nightly_tests == 'true' && '(nightly tests)' || '' }}
+            target: all
+    name: ${{ matrix.compiler_mode }} ${{ matrix.build_mode }} ${{ matrix.target }} ${{ matrix.tests_function }}
     runs-on: ubuntu-latest
     env:
       CC:         ${{ startsWith(matrix.compiler_mode, 'gcc') && 'gcc' || 'clang' }}
       LD:         ${{ startsWith(matrix.compiler_mode, 'gcc') && 'gcc' || 'clang' }}
       BUILD_MODE: ${{ matrix.build_mode }}
       BUILD_ASAN: ${{ endsWith(matrix.compiler_mode, 'asan') && 1 || 0 }}
       BUILD_MSAN: ${{ endsWith(matrix.compiler_mode, 'msan') && 1 || 0 }}
-      INCLUDE_SLOW_TESTS: ${{ matrix.include_slow_tests }}
-      RUN_NIGHTLY_TESTS: ${{ matrix.run_nightly_tests }}
+      TESTS_FUNCTION: ${{ matrix.tests_function }}
     steps:
     - name: Maximize build space
-      uses: easimon/maximize-build-space@master
+      uses: rtjohnso/maximize-build-space@master
       with:
-        root-reserve-mb: 2048
+        root-reserve-mb: 4096
+        temp-reserve-mb: 4096
         swap-size-mb: 1
         remove-dotnet: true
         remove-android: true
@@ -90,4 +76,4 @@ jobs:
       with:
         packages: libxxhash-dev libaio-dev libconfig-dev gcc clang-19 clang-format-19
     - name: make ${{ matrix.target }}
-      run: pwd; make ${{ matrix.target }}
+      run: df -h; make ${{ matrix.target }}

Makefile

@@ -558,7 +558,7 @@ run-tests: all-tests
 	BINDIR=$(BINDIR) ./test.sh
 
 test-results: all-tests
-	INCLUDE_SLOW_TESTS=true BINDIR=$(BINDIR) ./test.sh 2>&1 | tee ./test-results
+	BINDIR=$(BINDIR) ./test.sh 2>&1 | tee ./test-results
 
 run-examples: all-examples
 		for i in $(EXAMPLES_BINS); do $$i || exit; done

ci/steps.lib.yml

@@ -57,8 +57,6 @@ config:
   run:
     path: sh
     args: ["-c", "cd /splinterdb && ./test.sh"]
-  params:
-    INCLUDE_SLOW_TESTS: "true"
 #@ end
 
 ---
@@ -96,8 +94,6 @@ config:
   params:
     CC: #@ compiler
     LD: #@ compiler
-    INCLUDE_SLOW_TESTS: #@ str(not quick).lower()
-    RUN_NIGHTLY_TESTS: #@ str(test_nightly).lower()
     BUILD_VERBOSE: "1"
     VERBOSE: "3"
 

docs/testing.md

@@ -7,28 +7,16 @@ _Audience: All users and contributors._
 
 
 ## Testing Overview
-In CI, we execute these tests against the following build modes
-- All tests on optimized and debug builds using clang
-- All tests on optimized and debug builds using gcc
-- All tests using address-sanitizer debug builds using the clang compiler
-- All tests using memory-sanitizer debug builds using the gcc compiler
-- clang-format checks
-- [shellcheck](https://www.shellcheck.net) and shfmt checks run against shell scripts
-
-To run a small collection of unit-tests to give you a very quick
-baseline stability of the library, do: `$ make run-tests`
+To run the full test-suite, do: `$ make run-tests` or `$make test-results`.  The former runs the tests and outputs the results to the terminal.  The `test-results` target generates a file with the test results.
 
 The [`make run-tests`](../Makefile#:~:text=run%2Dtests) target invokes the
 underlying [`test.sh`](../test.sh) script to run quick tests.
 
-To execute a larger set of tests, including functional and performance tests,
-you can do one of the following:
-
+To execute a smaller set of tests, you can do the following:
 ```shell
-$ make test-results
-$ INCLUDE_SLOW_TESTS=true make run-tests
-$ INCLUDE_SLOW_TESTS=true ./test.sh
+$ TESTS_FUNCTION=<func> make <test-results | run-tests>
 ```
+See `test.sh` for valid values for `TESTS_FUNCTION`.
 
 In CI, all test execution is driven by the top-level [test.sh](../test.sh)
 script, which exercises individual build artifacts produced for testing, as

docs/unit-testing.md

@@ -73,14 +73,6 @@ unit-test binary with the `--list` argument.
 ----
 ## How To Run unit-tests
 
-Quick **smoke-tests** run of unit tests: `$ make run-tests`
-
-This runs a small subset of unit tests, which execute very quickly, so that you
-get a quick turnaround on the overall stability of your changes.
-
-You can achieve the same result by executing the following test driver script
-[test.sh](../test.sh#:~:text=if%20\[%20"$INCLUDE%5FSLOW%5FTESTS"%20!=%20"true"%20\]): `$ ./test.sh`
-
 To **run** all the unit-test suites: `$ bin/unit_test`
 
 To run a specific suite, optionally filtering on a specific test case:

include/splinterdb/splinterdb.h

@@ -34,16 +34,6 @@ splinterdb_get_version();
  * for all structures (default), or choose to setup a shared segment
  * which will be used for shared structures.
  *
- * ******************* EXPERIMENTAL FEATURES ********************
- *
- * - use_shmem: Support for shared memory segments:
- *   This flag will configure a shared memory segment. All (most) run-time
- *   memory allocation will be done from this shared segment. Currently,
- *   we do not support free(), so you will likely run out of shared memory
- *   and run into shared-memory OOM errors. This functionality is
- *   solely meant for internal development uses.
- *
- * ******************* EXPERIMENTAL FEATURES ********************
  */
 typedef struct splinterdb_config {
    // required configuration

src/clockcache.c

@@ -1223,9 +1223,7 @@ clockcache_get_free_page(clockcache *cc,
     * not give up after 3 passes on the cache. At least wait for the
     * max latency of an IO and keep making passes.
     */
-   while (num_passes < 3
-          || (blocking && !io_max_latency_elapsed(cc->io, wait_start)))
-   {
+   while (num_passes < 3 || blocking) {
       uint64 start_entry = cc->per_thread[tid].free_hand * CC_ENTRIES_PER_BATCH;
       uint64 end_entry   = start_entry + CC_ENTRIES_PER_BATCH;
       for (entry_no = start_entry; entry_no < end_entry; entry_no++) {

src/platform_linux/platform_io.h

@@ -37,7 +37,6 @@ struct iovec;
 #define IO_DEFAULT_FLAGS             (O_RDWR | O_CREAT)
 #define IO_DEFAULT_PERMS             (0755)
 #define IO_DEFAULT_KERNEL_QUEUE_SIZE (256)
-#define IO_DEFAULT_FILENAME          "db"
 #define IO_DEFAULT_ASYNC_QUEUE_DEPTH (256)
 
 /*

src/shard_log.c

@@ -227,8 +227,10 @@ shard_log_write(log_handle *logh, key tuple_key, message msg, uint64 generation)
       page        = cache_get(cc, thread_data->addr, TRUE, PAGE_TYPE_LOG);
       uint64 wait = 1;
       while (!cache_try_claim(cc, page)) {
+         cache_unget(cc, page);
          platform_sleep_ns(wait);
          wait = wait > 1024 ? wait : 2 * wait;
+         page = cache_get(cc, thread_data->addr, TRUE, PAGE_TYPE_LOG);
       }
       cache_lock(cc, page);
    }