feat: now compatible with non-local postgres (#110)

**Summary**: Modified code so that it can work with a non-local (meaning
on a different machine or in a Docker container) Postgres instance by
removing all uses of `pg_ctl` and `psql`.

**Demo**:
The demo shows that the codebase has no instances of "pg_ctl" or "psql"
(except for one appearance of "psql" in a comment). Despite this, `cargo
run ... cardtest` still works correctly.


https://github.com/cmu-db/optd/assets/20631215/453a341a-8141-4ac4-95f1-e09b7bb690a5

**Details**:
* Removing `pg_ctl` required significant changes. Instead of having one
`pgdata/` per TPC-H config's data, we now have one global `pgdata/` and
one _Postgres database_ per TPC-H config's data.
* Removing `psql` involved using SQL queries to load the database
instead of `psql` invocations.
* Since each TPC-H config corresponds to a Postgres database instead of
a file, the string representing each TPC-H config now has additional
restrictions described in `get_dbname()`.

Author: wangpatrick57

optd-perftest/Cargo.toml

@@ -24,4 +24,4 @@ clap = { version = "4.5", features = [
     "derive",
 ] }
 log = "0.4"
-env_logger = "0.11"
+env_logger = "0.11"

optd-perftest/src/benchmark.rs

@@ -7,11 +7,32 @@ pub enum Benchmark {
 }
 
 impl Benchmark {
-    pub fn get_stringid(&self) -> String {
-        match self {
+    /// Get a unique string that deterministically describes the "data" of this benchmark.
+    /// Note that benchmarks consist of "data" and "queries". This name is only for the data
+    /// For instance, if you have two TPC-H benchmarks with the same scale factor and seed
+    ///   but different queries, they could both share the same database and would thus
+    ///   have the same dbname.
+    /// This name must be compatible with the rules all databases have for their names, which
+    ///   are described below:
+    ///
+    /// Postgres' rules:
+    ///   - The name can only contain A-Z a-z 0-9 _ and cannot start with 0-9.
+    ///   - There is a weird behavior where if you use CREATE DATABASE to create a database,
+    ///     Postgres will convert uppercase letters to lowercase. However, if you use psql to
+    ///     then connect to the database, Postgres will *not* convert capital letters to
+    ///     lowercase. To resolve the inconsistency, the names output by this function will
+    ///     *not* contain uppercase letters.
+    pub fn get_dbname(&self) -> String {
+        let dbname = match self {
             Self::Test => String::from("test"),
-            Self::Tpch(tpch_config) => format!("tpch_{}", tpch_config.get_stringid()),
-        }
+            Self::Tpch(tpch_config) => {
+                format!("tpch_sf{}_sd{}", tpch_config.scale_factor, tpch_config.seed)
+            }
+        };
+        // since Postgres names cannot contain periods
+        let dbname = dbname.replace('.', "point");
+        // due to the weird inconsistency with Postgres (see function comment)
+        dbname.to_lowercase()
     }
 
     pub fn is_readonly(&self) -> bool {

optd-perftest/src/cardtest.rs

@@ -77,15 +77,6 @@ pub trait CardtestRunnerDBHelper {
     fn get_name(&self) -> &str;
 
     // The order of queries has to be the same between these two functions.
-    // They take mutable references because evaluation sometimes involves mutating self.
-    //   One example of this is in PostgresDb where we may need to reconnect to the database,
-    //   which requires modifying the PostgresDb object.
-    async fn eval_benchmark_estcards(
-        &mut self,
-        benchmark: &Benchmark,
-    ) -> anyhow::Result<Vec<usize>>;
-    async fn eval_benchmark_truecards(
-        &mut self,
-        benchmark: &Benchmark,
-    ) -> anyhow::Result<Vec<usize>>;
+    async fn eval_benchmark_estcards(&self, benchmark: &Benchmark) -> anyhow::Result<Vec<usize>>;
+    async fn eval_benchmark_truecards(&self, benchmark: &Benchmark) -> anyhow::Result<Vec<usize>>;
 }

optd-perftest/src/datafusion_db_cardtest.rs

@@ -8,17 +8,11 @@ impl CardtestRunnerDBHelper for DatafusionDb {
         "DataFusion"
     }
 
-    async fn eval_benchmark_truecards(
-        &mut self,
-        _benchmark: &Benchmark,
-    ) -> anyhow::Result<Vec<usize>> {
+    async fn eval_benchmark_truecards(&self, _benchmark: &Benchmark) -> anyhow::Result<Vec<usize>> {
         Ok(vec![])
     }
 
-    async fn eval_benchmark_estcards(
-        &mut self,
-        _benchmark: &Benchmark,
-    ) -> anyhow::Result<Vec<usize>> {
+    async fn eval_benchmark_estcards(&self, _benchmark: &Benchmark) -> anyhow::Result<Vec<usize>> {
         Ok(vec![])
     }
 }

optd-perftest/src/main.rs

@@ -66,7 +66,7 @@ async fn cardtest<P: AsRef<Path>>(
     workspace_dpath: P,
     tpch_config: TpchConfig,
 ) -> anyhow::Result<()> {
-    let pg_db = PostgresDb::build(workspace_dpath).await?;
+    let pg_db = PostgresDb::new(workspace_dpath);
     let databases: Vec<Box<dyn CardtestRunnerDBHelper>> = vec![Box::new(pg_db)];
 
     let tpch_benchmark = Benchmark::Tpch(tpch_config.clone());