Merge branch 'deprecate-clickhouse-tcp' of github.com:sundy-li/fuse-q…

…uery into deprecate-clickhouse-tcp

.github/actions/artifact_failure/action.yml

@@ -18,6 +18,6 @@ runs:
           query/_logs*/
           store/_logs*/
           .databend/
-          query.log
+          query*.log
           metasrv.log
           nohup.out

.github/actions/fuse_compat/action.yml

@@ -0,0 +1,36 @@
+name: "Test fuse-table format in an old query is compatible with new query"
+description: "Download old binaries and current binaries, write data with old query, read data with new query"
+inputs:
+  profile:
+    description: ""
+    required: true
+    default: "debug"
+  target:
+    description: ""
+    required: true
+    default: "x86_64-unknown-linux-gnu"
+runs:
+  using: "composite"
+  steps:
+    - name: Setup Build Tool
+      uses: ./.github/actions/setup_build_tool
+
+    - name: Download artifact
+      uses: ./.github/actions/artifact_download
+      with:
+        profile: ${{ inputs.profile }}
+        sha: ${{ github.sha }}
+        target: ${{ inputs.target }}
+        path: ./bins/current
+
+    - name: Test compatibility
+      shell: bash
+      run: |
+        build-tool bash ./tests/fuse-compat/test-fuse-compat.sh 0.7.150
+        build-tool bash ./tests/fuse-compat/test-fuse-compat.sh 0.7.151
+
+    - name: Upload failure
+      if: failure()
+      uses: ./.github/actions/artifact_failure
+      with:
+        name: test-compat

.github/workflows/dev-linux.yml

@@ -96,6 +96,14 @@ jobs:
       - uses: actions/checkout@v3
       - uses: ./.github/actions/test_compat
 
+  fuse_compat:
+    timeout-minutes: 10
+    runs-on: [self-hosted, X64, Linux, development]
+    needs: build_gnu
+    steps:
+      - uses: actions/checkout@v3
+      - uses: ./.github/actions/fuse_compat
+
   test_meta_cluster:
     timeout-minutes: 10
     runs-on: [self-hosted, X64, Linux, development]

query/src/sql/optimizer/heuristic/prune_columns.rs

@@ -116,7 +116,13 @@ impl ColumnPruner {
                 ))
             }
             RelOperator::Project(p) => {
-                let used: ColumnSet = p.columns.intersection(&required).cloned().collect();
+                let mut used: ColumnSet = p.columns.intersection(&required).cloned().collect();
+                if used.is_empty() {
+                    // Keep at least one column for project.
+                    used.insert(*p.columns.iter().sorted().take(1).next().ok_or_else(|| {
+                        ErrorCode::LogicalError("Invalid Project without output column")
+                    })?);
+                }
                 Ok(SExpr::create_unary(
                     RelOperator::Project(Project {
                         columns: used.clone(),

query/tests/it/sql/optimizer/heuristic/prune_columns.rs

@@ -60,6 +60,11 @@ pub async fn test_heuristic_optimizer_prune_columns() -> Result<()> {
             query: "select name from system.functions order by example".to_string(),
             rules: vec![],
         },
+        Suite {
+            comment: "# Prune unused columns with cross join".to_string(),
+            query: "select t.number from numbers(10) t where exists(select * from numbers(10))".to_string(),
+            rules: vec![],
+        },
     ];
 
     run_suites(ctx, &mut file, &suites, run_test).await

query/tests/it/sql/optimizer/heuristic/testdata/prune_columns.test

@@ -88,3 +88,20 @@ Project: [name (#0)]
             Scan: default.system.functions
 
 
+# Prune unused columns with cross join
+select t.number from numbers(10) t where exists(select * from numbers(10))
+----
+Project: [number (#0)]
+    EvalScalar: [t.number (#0)]
+        Filter: [subquery_3 (#3)]
+            CrossJoin
+                Scan: default.system.numbers
+                Project: [subquery (#3)]
+                    EvalScalar: [count(*) (#2) = 1]
+                        Aggregate(Initial): group items: [], aggregate functions: [count(*)]
+                            Limit: [1], Offset: [0]
+                                Project: [number (#1)]
+                                    EvalScalar: [numbers.number (#1)]
+                                        Scan: default.system.numbers
+
+

query/tests/it/sql/optimizer/heuristic/testdata/subquery.test

@@ -36,7 +36,7 @@ select t.number from numbers(1) as t where exists (select * from numbers(1) wher
 Project: [number (#0)]
     CrossJoin
         Scan: default.system.numbers
-        Project: []
+        Project: [subquery (#3)]
             Filter: [subquery_3 (#3)]
                 EvalScalar: [count(*) (#2) = 1]
                     Aggregate(Final): group items: [], aggregate functions: [count(*)]

tests/fuse-compat/README.md

@@ -0,0 +1,31 @@
+# Fuse table compatability test
+
+This script tests whether a newer version databend-query can read fuse table data written
+by a older version databend-query.
+
+## Usage
+
+```shell
+tests/fuse-compat/test-fuse-compat.sh <old_ver>
+```
+
+E.g. `tests/fuse-compat/test-fuse-compat.sh 0.7.151` tests if the fuse-table written
+by **databend-query-0.7.151** can be read by **current** version databend-qeury.
+
+## Prerequisites
+
+- Current version of databend-query and databend-meta must reside in `./bins`:
+    - `./bins/current/databend-query`
+    - `./bins/current/databend-meta`
+
+    Since building a binary takes several minutes,
+    this step is usually done by the calling process, e.g., the CI script.
+
+
+## Testing data
+
+- Suite `tests/fuse-compat/compat-logictest/fuse_compat_write` writes data into a fuse table via an old version query.
+- Suite `tests/fuse-compat/compat-logictest/fuse_compat_read` reads the data via current version query.
+
+Fuse table maintainers update these two `logictest` scripts to let the write/read
+operations cover fuse-table features.

tests/fuse-compat/compat-logictest/fuse_compat_read

@@ -0,0 +1,6 @@
+statement query I
+select a+b from t;
+
+----
+2
+4

tests/fuse-compat/compat-logictest/fuse_compat_write

@@ -0,0 +1,8 @@
+statement ok
+DROP TABLE IF EXISTS t;
+
+statement ok
+create table t(a int,b int) Engine = Fuse;
+
+statement ok
+insert into t values(1,1),(2,2);

tests/fuse-compat/test-fuse-compat.sh

@@ -0,0 +1,228 @@
+#!/bin/sh
+
+set -o errexit
+SCRIPT_PATH="$(cd "$(dirname "$0")" >/dev/null 2>&1 && pwd)"
+echo " === SCRIPT_PATH: $SCRIPT_PATH"
+# go to work tree root
+cd "$SCRIPT_PATH/../../"
+ROOT="$(pwd)"
+pwd
+
+BUILD_PROFILE="${BUILD_PROFILE:-debug}"
+
+query_config_path="scripts/ci/deploy/config/databend-query-node-1.toml"
+logictest_path="tests/fuse-compat/compat-logictest"
+bend_repo_url="https://github.com/datafuselabs/databend"
+
+usage() {
+    echo " === Assert that latest query being compatible with an old version query on fuse-table format"
+    echo " === Expect ./bins/current contains current version binaries"
+    echo " === Usage: $0 <old_version>"
+}
+
+binary_url() {
+    local ver="$1"
+    echo "https://github.com/datafuselabs/databend/releases/download/v${ver}-nightly/databend-v${ver}-nightly-x86_64-unknown-linux-gnu.tar.gz"
+}
+
+# download a specific version of databend, untar it to folder `./bins/$ver`
+# `ver` is semver without prefix `v` or `-nightly`
+download_binary() {
+    local ver="$1"
+
+    local url="$(binary_url $ver)"
+    local fn="databend-$ver.tar.gz"
+
+    if [ -f ./bins/$ver/databend-query ]; then
+        echo " === binaries exist: $(ls ./bins/$ver/* | tr '\n' ' ')"
+        chmod +x ./bins/$ver/*
+        return
+    fi
+
+    if [ -f "$fn" ]; then
+        echo " === tar file exists: $fn"
+    else
+        echo " === Download binary ver: $ver"
+        echo " === Download binary url: $url"
+
+        curl -L "$url" >"$fn"
+        # or:
+        # wget -q "$url" -o "$fn"
+    fi
+
+    mkdir -p ./bins/$ver
+    tar -xf "$fn" -C ./bins/$ver
+
+    echo " === unpacked: ./bins/$ver:"
+    ls ./bins/$ver
+
+    chmod +x ./bins/$ver/*
+}
+
+# Clone only specified dir or file in the specified commit
+git_partial_clone() {
+    local repo_url="$1"
+    local branch="$2"
+    local worktree_path="$3"
+    local local_path="$4"
+
+    echo " === Clone $repo_url@$branch:$worktree_path"
+    echo " ===    To $local_path/$worktree_path"
+
+    rm -rf "$local_path" || echo "no $local_path"
+
+    git clone \
+        -b "$branch" \
+        --depth 1 \
+        --filter=blob:none \
+        --sparse \
+        "$repo_url" \
+        "$local_path"
+
+    cd "$local_path"
+    git sparse-checkout set "$worktree_path"
+
+    echo " === Done clone from $repo_url@$branch:$worktree_path"
+
+    ls "$worktree_path"
+
+}
+
+
+# Run specified tests found in logic test suite dir
+run_logictest()
+{
+    local pattern="$1"
+
+    (
+        cd "tests/logictest"
+        python3 main.py --suites "$SCRIPT_PATH/compat-logictest" "$pattern"
+    )
+
+}
+
+
+kill_proc() {
+    local name="$1"
+
+    echo " === Kill $name ..."
+
+    killall "$name" || {
+        echo " === no "$name" to kill"
+        return
+    }
+
+    sleep 1
+
+    if test -n "$(pgrep $name)"; then
+        echo " === The $name is not killed. force killing."
+        killall -9 $name || echo " === no $Name to killall-9"
+    fi
+
+    echo " === Done kill $name"
+}
+
+# Test fuse-data compatibility between an old version query and the current
+# version query.
+run_test() {
+    local query_old_ver="$1"
+
+    echo " === Test with query-$query_old_ver and current query"
+
+    local query_old="./bins/$query_old_ver/bin/databend-query"
+    local query_new="./bins/current/databend-query"
+    local metasrv="./bins/current/databend-meta"
+
+
+    echo " === metasrv version:"
+    # TODO remove --single
+    "$metasrv" --single --cmd ver || echo " === no version yet"
+
+    echo " === old query version:"
+    "$query_old" --cmd ver || echo " === no version yet"
+
+    echo " === new query version:"
+    "$query_new" --cmd ver || echo " === no version yet"
+
+    sleep 1
+
+
+    kill_proc databend-query
+    kill_proc databend-meta
+
+    echo " === Clean meta dir"
+    rm -rf .databend || echo " === no dir to rm: .databend"
+
+    rm nohup.out || echo "no nohup.out"
+
+
+    echo ' === Start databend-meta...'
+
+    export RUST_BACKTRACE=1
+
+    nohup "$metasrv" --single --log-level=DEBUG &
+    python3 scripts/ci/wait_tcp.py --timeout 5 --port 9191
+
+
+    # Only run test on mysql handler
+    export DISABLE_HTTP_LOGIC_TEST=true
+    export DISABLE_CLICKHOUSE_LOGIC_TEST=true
+
+
+    echo ' === Start old databend-query...'
+
+    # (
+    #     download_query_config "$query_old_ver" old_config/$query_old_ver
+    # )
+    # config_path="old_config/$query_old_ver/$query_config_path"
+    config_path="./bins/$query_old_ver/configs/databend-query.toml"
+
+
+    # TODO clean up data?
+    echo " === bring up $query_old"
+
+    nohup "$query_old" -c "$config_path" --log-level DEBUG --meta-endpoints "0.0.0.0:9191" >query-old.log &
+    python3 scripts/ci/wait_tcp.py --timeout 5 --port 3307
+
+    echo " === Run test: fuse_compat_write with old query"
+
+    # download_logictest $query_old_ver old_logictest/$query_old_ver
+    # run_logictest old_logictest/$query_old_ver fuse_compat_write
+    run_logictest fuse_compat_write
+
+
+    kill_proc databend-query
+
+
+    echo " === bring up new query "
+
+    config_path="$query_config_path"
+
+    nohup "$query_new" -c "$config_path" --log-level DEBUG --meta-endpoints "0.0.0.0:9191" >query-current.log &
+    python3 scripts/ci/wait_tcp.py --timeout 5 --port 3307
+
+
+    echo " === Run test: fuse_compat_read with current query"
+
+    # download_logictest $query_old_ver old_logictest
+    run_logictest fuse_compat_read
+}
+
+# -- main --
+
+# The previous version to assert compatibility with
+# e.g. old_query_ver="0.7.151"
+old_query_ver="$1"
+
+
+chmod +x ./bins/current/*
+
+echo " === current metasrv ver: $(./bins/current/databend-meta --single --cmd ver | tr '\n' ' ')"
+echo " === current   query ver: $(./bins/current/databend-query --cmd ver | tr '\n' ' ')"
+echo " === old query ver: $old_query_ver"
+
+download_binary "$old_query_ver"
+
+mkdir -p ./target/${BUILD_PROFILE}/
+
+run_test $old_query_ver