Question: Do we need any logic on shutdown?

[ceckoslab]

I have been playing with chdb-node a lot lately - exploring what can I do with the project. I am working on a proof of concept a local emulator of a cloud based datawarehouse.

I programmed a lot logic in a single index.js file and I was running and stopping the index.js a few times a day.

Today I got a strange error after is was trying to start again the index.js with recent changes:

Attempt 1:

[1]    91711 abort      node index.js

Attempt 2::

node(91807,0x174d13000) malloc: *** error for object 0x6000001f4180: pointer being freed was not allocated
node(91807,0x174d13000) malloc: *** set a breakpoint in malloc_error_break to debug
node(91807,0x17470f000) malloc: *** error for object 0x600003a968c0: pointer being freed was not allocated
node(91807,0x17470f000) malloc: *** set a breakpoint in malloc_error_break to debug

After deleting the chdb data folder everything started working fine.

Here comes the question:

We we need to do or can we do something on SIGINT and SIGTERM ? I suppose that there could be cases where we could cause data corruption if we do not stop properly chdb while chdb is still processing data.

For example I noticed this code fragment in another project that takes requests and writes them to file system:

  // Handle close event
  process
    .once("message", (msg) => {
      if (msg === "shutdown") {
        shutdown(server);
      }
    })
    .once("SIGINT", () => shutdown(server))
    .once("SIGTERM", () => shutdown(server));
}

Full source code here: https://github.com/Azure/Azurite/blob/76f626284e4b4b58b95065bb3c92351f30af7f3d/src/blob/main.ts

[ceckoslab]

@auxten I was able to create a repro in a single file.

Here is an example:

const { query, Session } = require("chdb");

var ret;

// Test standalone query
ret = query("SELECT version(), 'Hello chDB', chdb()", "CSV");
console.log("Standalone Query Result:", ret);

// Test session query
// Create a new session instance
const session = new Session("./corrupt-on-shutdown");

session.query("CREATE DATABASE IF NOT EXISTS testdb;");

session.query("USE testdb;");

// Let's create a bunch of talbes
for (let t = 0; t < 20; t++) {
    console.log("Running create table num: " + t);
    session.query(`CREATE TABLE IF NOT EXISTS testtable_${t} (id UInt32) ENGINE = MergeTree() ORDER BY id;`)
}

// Let's insert bunch of data
for (let c = 0; c < 30; c++) {
    console.log("Running query num: " + c);
    session.query("INSERT INTO testtable_0 VALUES (1), (2), (3);");
} 

ret = session.query("SELECT * FROM testtable_0;")
console.log("Session Query Result:", ret);

// Handle the SIGINT signal (Ctrl+C)
process.on('SIGINT', () => {
    console.log("\nReceived SIGINT. Exiting gracefully...");
    process.exit(0);  // Exit with a success code
});

// Keep the script running indefinitely
console.log("Press Ctrl+C to stop the script.");
setInterval(() => {}, 1000);  // An empty interval to keep the event loop active

I get the error on every 2-3 runs.

I am waiting till I see the message "Press Ctrl+C to stop the script." and I press Ctrl+C few seconds after that.

System info:

• Apple M2 Pro
• macOS 14.5 (23F79)
• Node v20.9.0

  "dependencies": {
    "chdb": "^1.2.1"
  },

[auxten]

I should fix this issue first chdb-io/chdb#197

[ahawker]

Any action on this? I was seeing the same in golang multiple months back and basically made it a blocker for any long running process trying to embed/use chdb.

[auxten]

It seems to be the chDB c++ part of the code that captures the signal, causing the language itself to be unable to process the signal. We are looking at this problem and hope it can be solved by next week.

[auxten]

Here is the plan:

1. Disable the singal handlers in libchdb, Signal handling function in chdb's C++ component blocks long-running apps from setting signal handlers chdb#321
2. Update chdb-node to libchdb v3 which has better support for stateful usage

[ahawker]

Thanks! Looking forward to testing this out in the next release.