Merge pull request #14838 from serathius/linearizability-docs

tests: Document robustness tests
2023-03-28 16:22:09 +02:00 · 2023-03-28 16:22:09 +02:00 · 5223d09d41
parent 8a3ffbd7a9 d03ac88b36
commit 5223d09d41
7 changed files with 173 additions and 53 deletions
--- a/.github/workflows/robustness-template.yaml
+++ b/.github/workflows/robustness-template.yaml
@ -33,18 +33,19 @@ jobs:

        case "${GITHUB_REF}" in
          release-3.5)
-            make build-failpoints-release-3.5
-            ./bin/etcd --version
+            make /tmp/etcd-release-3.5-failpoints/bin/etcd
+            cp /tmp/etcd-release-3.5-failpoints/bin/etcd bin/etcd
            ;;
          release-3.4)
-            make build-failpoints-release-3.4
-            ./bin/etcd --version
+            make /tmp/etcd-release-3.4-failpoints/bin/etcd
+            cp /tmp/etcd-release-3.4-failpoints/bin/etcd bin/etcd
            ;;
          *)
            make gofail-enable
            make build
            ;;
          esac
+        ./bin/etcd --version
    - name: test-robustness
      run: |
        set -euo pipefail
--- a/50
+++ b/50
@ -1,3 +1,6 @@
+all: build
+include tests/robustness/makefile.mk
+
 .PHONY: build
 build:
 	GO_BUILD_FLAGS="${GO_BUILD_FLAGS} -v" ./scripts/build.sh
@ -113,53 +116,6 @@ verify-genproto:
 verify-goimport:
 	PASSES="goimport" ./scripts/test.sh

-# Failpoints
-
-GOFAIL_VERSION = $(shell cd tools/mod && go list -m -f {{.Version}} go.etcd.io/gofail)
-
-.PHONY: gofail-enable
-gofail-enable: install-gofail
-	gofail enable server/etcdserver/ server/storage/backend/ server/storage/mvcc/ server/storage/wal/
-	cd ./server && go get go.etcd.io/gofail@${GOFAIL_VERSION}
-	cd ./etcdutl && go get go.etcd.io/gofail@${GOFAIL_VERSION}
-	cd ./etcdctl && go get go.etcd.io/gofail@${GOFAIL_VERSION}
-	cd ./tests && go get go.etcd.io/gofail@${GOFAIL_VERSION}
-
-.PHONY: gofail-disable
-gofail-disable: install-gofail
-	gofail disable server/etcdserver/ server/storage/backend/ server/storage/mvcc/ server/storage/wal/
-	cd ./server && go mod tidy
-	cd ./etcdutl && go mod tidy
-	cd ./etcdctl && go mod tidy
-	cd ./tests && go mod tidy
-
-.PHONY: install-gofail
-install-gofail:
-	cd tools/mod; go install go.etcd.io/gofail@${GOFAIL_VERSION}
-
-build-failpoints-release-3.5:
-	rm -rf /tmp/etcd-release-3.5/
-	mkdir -p /tmp/etcd-release-3.5/
-	cd /tmp/etcd-release-3.5/; \
-	  git clone --depth 1 --branch release-3.5 https://github.com/etcd-io/etcd.git .; \
-	  go get go.etcd.io/gofail@${GOFAIL_VERSION}; \
-	  (cd server; go get go.etcd.io/gofail@${GOFAIL_VERSION}); \
-	  (cd etcdctl; go get go.etcd.io/gofail@${GOFAIL_VERSION}); \
-	  (cd etcdutl; go get go.etcd.io/gofail@${GOFAIL_VERSION}); \
-	  FAILPOINTS=true ./build;
-	mkdir -p ./bin
-	cp /tmp/etcd-release-3.5/bin/etcd ./bin/etcd
-
-build-failpoints-release-3.4:
-	rm -rf /tmp/etcd-release-3.4/
-	mkdir -p /tmp/etcd-release-3.4/
-	cd /tmp/etcd-release-3.4/; \
-	  git clone --depth 1 --branch release-3.4 https://github.com/etcd-io/etcd.git .; \
-	  go get go.etcd.io/gofail@${GOFAIL_VERSION}; \
-	  FAILPOINTS=true ./build;
-	mkdir -p ./bin
-	cp /tmp/etcd-release-3.4/bin/etcd ./bin/etcd
-
 # Cleanup

 clean:
--- a/scripts/test.sh
+++ b/scripts/test.sh
@ -129,7 +129,7 @@ function e2e_pass {

 function robustness_pass {
  # e2e tests are running pre-build binary. Settings like --race,-cover,-cpu does not have any impact.
-  run_for_module "tests" go_test "./robustness/..." "keep_going" : -timeout="${TIMEOUT:-30m}" "${RUN_ARG[@]}" "$@"
+  run_for_module "tests" go_test "./robustness" "keep_going" : -timeout="${TIMEOUT:-30m}" "${RUN_ARG[@]}" "$@"
 }

 function integration_e2e_pass {
--- a/scripts/test_lib.sh
+++ b/scripts/test_lib.sh
@ -315,7 +315,7 @@ function go_test {
    additional_flags=$(${flags_for_package_func} ${pkg})

    # shellcheck disable=SC2206
-    local cmd=( go test ${goTestFlags} ${additional_flags} "$@" ${pkg} )
+    local cmd=( go test ${goTestFlags} ${additional_flags} ${pkg} "$@" )

    # shellcheck disable=SC2086
    if ! run env ${goTestEnv} ETCD_VERIFY="${ETCD_VERIFY}" "${cmd[@]}" | tee ${junit_filename_prefix:+"${junit_filename_prefix}.stdout"} | grep --binary-files=text "${go_test_grep_pattern}" ; then
--- a/tests/robustness/README.md
+++ b/tests/robustness/README.md
@ -0,0 +1,71 @@
+# etcd Robustness Testing
+
+Purpose of etcd robustness tests is to validate that etcd upholds
+[API guarantees] and [watch guarantees] under any condition or failure.
+
+Robustness tests achieve that comparing etcd cluster behavior against a simplified model.
+Multiple test encompass different etcd cluster setups, client traffic types and failures experienced by cluster.
+During a single test we create a cluster and inject failures while sending and recording client traffic.
+Correctness is validated by running collected history of client operations against the etcd model and a set of validators.
+Upon failure tests generate a report that can be used to attribute whether failure was caused by bug in etcd or test framework. 
+
+[API guarantees]: https://etcd.io/docs/latest/learning/api_guarantees/
+[watch guarantees]: https://etcd.io/docs/latest/learning/api/#watch-streams
+
+## Running locally 
+
+1. Build etcd with failpoints
+    ```bash
+    make gofail-enable
+    make build
+    make gofail-disable
+    ```
+2. Run the tests
+
+    ```bash
+    make test-robustness
+    ```
+   
+    Optionally you can pass environment variables:
+    * `GO_TEST_FLAGS` - to pass additional arguments to `go test`. 
+      It is recommended to run tests multiple times with failfast enabled. this can be done by setting `GO_TEST_FLAGS='--count=100 --failfast'`.
+    * `EXPECT_DEBUG=true` - to get logs from the cluster.
+    * `RESULTS_DIR` - to change location where results report will be saved.
+
+## Analysing failure
+
+If robustness tests fails we want to analyse the report to confirm if the issue is on etcd side. Location of this report
+is included in test logs. One of log lines should look like:
+```
+    history.go:34: Model is not linearizable
+    logger.go:130: 2023-03-18T12:18:03.244+0100 INFO    Saving member data dir  {"member": "TestRobustnessIssue14370-test-0", "path": "/tmp/TestRobustness_Issue14370/TestRobustnessIssue14370-test-0"}
+    logger.go:130: 2023-03-18T12:18:03.244+0100 INFO    Saving watch responses  {"path": "/tmp/TestRobustness_Issue14370/TestRobustnessIssue14370-test-0/responses.json"}
+    logger.go:130: 2023-03-18T12:18:03.247+0100 INFO    Saving watch events     {"path": "/tmp/TestRobustness_Issue14370/TestRobustnessIssue14370-test-0/events.json"}
+    logger.go:130: 2023-03-18T12:18:03.248+0100 INFO    Saving operation history        {"path": "/tmp/TestRobustness_Issue14370/full-history.json"}
+    logger.go:130: 2023-03-18T12:18:03.252+0100 INFO    Saving operation history        {"path": "/tmp/TestRobustness_Issue14370/patched-history.json"}
+    logger.go:130: 2023-03-18T12:18:03.256+0100 INFO    Saving visualization    {"path": "/tmp/TestRobustness_Issue14370/history.html"}
+```
+
+Report includes multiple types of files:
+* Member db files, can be used to verify disk/memory corruption.
+* Watch responses saved as json, can be used to validate [watch guarantees].
+* Operation history saved as both html visualization and a json, can be used to validate [API guarantees].
+
+### Example analysis of linearization issue
+
+Let's analyse issue [#14370].
+To reproduce the issue by yourself run `make test-robustness-issue14370`.
+After a couple of tries robustness tests should report `Model is not linearizable` and save report locally.
+Lineralization issues are easiest to analyse via history visualization. 
+Open `/tmp/TestRobustness_Issue14370/history.html` file in your browser.
+Jump to the error in linearization by clicking `[ jump to first error ]` on the top of the page.
+
+You should see a graph similar to the one on the image below.
+![issue14370](./issue14370.png)
+
+Last correct request (connected with grey line) is a `Put` request that succeeded and got revision `168`.
+All following requests are invalid (connected with red line) as they have revision `167`. 
+Etcd guarantee that revision is non-decreasing, so this shows a bug in etcd as there is no way revision should decrease.
+This is consistent with the root cause of [#14370] as it was issue with process crash causing last write to be lost.
+
+[#14370]: https://github.com/etcd-io/etcd/issues/14370
--- a/tests/robustness/issue14370.png
+++ b/tests/robustness/issue14370.png
--- a/tests/robustness/makefile.mk
+++ b/tests/robustness/makefile.mk
@ -0,0 +1,92 @@
+# Reproduce historical issues
+
+.PHONY: test-robustness-issue14370
+test-robustness-issue14370: /tmp/etcd-v3.5.4-failpoints/bin
+	GO_TEST_FLAGS='-v --run=TestRobustness/Issue14370 --count 100 --failfast --bin-dir=/tmp/etcd-v3.5.4-failpoints/bin' make test-robustness && \
+	 echo "Failed to reproduce" || echo "Successful reproduction"
+
+.PHONY: test-robustness-issue13766
+test-robustness-issue13766: /tmp/etcd-v3.5.2-failpoints/bin
+	GO_TEST_FLAGS='-v --run=TestRobustness/Issue13766 --count 100 --failfast --bin-dir=/tmp/etcd-v3.5.2-failpoints/bin' make test-robustness && \
+	 echo "Failed to reproduce" || echo "Successful reproduction"
+
+.PHONY: test-robustness-issue14685
+test-robustness-issue14685: /tmp/etcd-v3.5.5-failpoints/bin
+	GO_TEST_FLAGS='-v --run=TestRobustness/Issue14685 --count 100 --failfast --bin-dir=/tmp/etcd-v3.5.5-failpoints/bin' make test-robustness && \
+	 echo "Failed to reproduce" || echo "Successful reproduction"
+
+# Failpoints
+
+GOFAIL_VERSION = $(shell cd tools/mod && go list -m -f {{.Version}} go.etcd.io/gofail)
+
+.PHONY: gofail-enable
+gofail-enable: install-gofail
+	gofail enable server/etcdserver/ server/storage/backend/ server/storage/mvcc/ server/storage/wal/
+	cd ./server && go get go.etcd.io/gofail@${GOFAIL_VERSION}
+	cd ./etcdutl && go get go.etcd.io/gofail@${GOFAIL_VERSION}
+	cd ./etcdctl && go get go.etcd.io/gofail@${GOFAIL_VERSION}
+	cd ./tests && go get go.etcd.io/gofail@${GOFAIL_VERSION}
+
+.PHONY: gofail-disable
+gofail-disable: install-gofail
+	gofail disable server/etcdserver/ server/storage/backend/ server/storage/mvcc/ server/storage/wal/
+	cd ./server && go mod tidy
+	cd ./etcdutl && go mod tidy
+	cd ./etcdctl && go mod tidy
+	cd ./tests && go mod tidy
+
+.PHONY: install-gofail
+install-gofail:
+	cd tools/mod; go install go.etcd.io/gofail@${GOFAIL_VERSION}
+
+# Build previous releases for robustness tests
+
+/tmp/etcd-v3.6.0-failpoints/bin:
+	rm -rf /tmp/etcd-v3.6.0-failpoints/
+	mkdir -p /tmp/etcd-v3.6.0-failpoints/
+	cd /tmp/etcd-v3.6.0-failpoints/; \
+	  git clone --depth 1 --branch main https://github.com/etcd-io/etcd.git .; \
+	  make gofail-enable; \
+	  make build;
+
+/tmp/etcd-v3.5.2-failpoints/bin:
+/tmp/etcd-v3.5.4-failpoints/bin:
+/tmp/etcd-v3.5.5-failpoints/bin:
+/tmp/etcd-v3.5.%-failpoints/bin:
+	rm -rf /tmp/etcd-v3.5.$*-failpoints/
+	mkdir -p /tmp/etcd-v3.5.$*-failpoints/
+	cd /tmp/etcd-v3.5.$*-failpoints/; \
+	  git clone --depth 1 --branch v3.5.$* https://github.com/etcd-io/etcd.git .; \
+	  go get go.etcd.io/gofail@${GOFAIL_VERSION}; \
+	  (cd server; go get go.etcd.io/gofail@${GOFAIL_VERSION}); \
+	  (cd etcdctl; go get go.etcd.io/gofail@${GOFAIL_VERSION}); \
+	  (cd etcdutl; go get go.etcd.io/gofail@${GOFAIL_VERSION}); \
+	  FAILPOINTS=true ./build;
+
+/tmp/etcd-release-3.5-failpoints/bin/etcd:
+	rm -rf /tmp/etcd-release-3.5-failpoints/
+	mkdir -p /tmp/etcd-release-3.5-failpoints/
+	cd /tmp/etcd-release-3.5-failpoints/; \
+	  git clone --depth 1 --branch release-3.5 https://github.com/etcd-io/etcd.git .; \
+	  go get go.etcd.io/gofail@${GOFAIL_VERSION}; \
+	  (cd server; go get go.etcd.io/gofail@${GOFAIL_VERSION}); \
+	  (cd etcdctl; go get go.etcd.io/gofail@${GOFAIL_VERSION}); \
+	  (cd etcdutl; go get go.etcd.io/gofail@${GOFAIL_VERSION}); \
+	  FAILPOINTS=true ./build;
+
+/tmp/etcd-v3.4.23-failpoints/bin:
+/tmp/etcd-v3.4.%-failpoints/bin:
+	rm -rf /tmp/etcd-v3.4.$*-failpoints/
+	mkdir -p /tmp/etcd-v3.4.$*-failpoints/
+	cd /tmp/etcd-v3.4.$*-failpoints/; \
+	  git clone --depth 1 --branch v3.4.$* https://github.com/etcd-io/etcd.git .; \
+	  go get go.etcd.io/gofail@${GOFAIL_VERSION}; \
+	  FAILPOINTS=true ./build;
+
+/tmp/etcd-release-3.4-failpoints/bin/etcd:
+	rm -rf /tmp/etcd-release-3.4-failpoints/
+	mkdir -p /tmp/etcd-release-3.4-failpoints/
+	cd /tmp/etcd-release-3.4-failpoints/; \
+	  git clone --depth 1 --branch release-3.4 https://github.com/etcd-io/etcd.git .; \
+	  go get go.etcd.io/gofail@${GOFAIL_VERSION}; \
+	  FAILPOINTS=true ./build;