[OSD Availability] Prevent OSD process crashes when disk is full

changelogs/fragments/6733.yml

@@ -0,0 +1,2 @@
+fix:
+- [OSD Availability] Prevent OSD process crashes when disk is full ([#6733](https://github.com/opensearch-project/OpenSearch-Dashboards/pull/6733))

config/opensearch_dashboards.yml

@@ -135,6 +135,11 @@
 # Enables you to specify a file where OpenSearch Dashboards stores log output.
 #logging.dest: stdout
 
+# This configuration option controls the handling of error messages in the logging stream. It is disabled by default.
+# When set to true, the 'ENOSPC' error message will not cause the OpenSearch Dashboards process to crash. Otherwise,
+# the original behavior will be maintained.
+#logging.ignoreEnospcError: false
+
 # Set the value of this setting to true to suppress all logging output.
 #logging.silent: false
 

src/dev/build/tasks/os_packages/docker_generator/resources/bin/opensearch-dashboards-docker

@@ -58,6 +58,7 @@ opensearch_dashboards_vars=(
     opensearchDashboards.defaultAppId
     opensearchDashboards.index
     logging.dest
+    logging.ignoreEnospcError
     logging.json
     logging.quiet
     logging.rotate.enabled

src/legacy/server/config/schema.js

@@ -109,6 +109,7 @@ export default () =>
         }),
         events: Joi.any().default({}),
         dest: Joi.string().default('stdout'),
+        ignoreEnospcError: Joi.boolean().default(false),
         filter: Joi.any().default({}),
         json: Joi.boolean().when('dest', {
           is: 'stdout',

src/legacy/server/logging/configuration.js

@@ -64,6 +64,7 @@ export default function loggingConfiguration(config) {
       json: config.get('logging.json'),
       dest: config.get('logging.dest'),
       timezone: config.get('logging.timezone'),
+      ignoreEnospcError: config.get('logging.ignoreEnospcError'),
 
       // I'm adding the default here because if you add another filter
       // using the commandline it will remove authorization. I want users

src/legacy/server/logging/log_reporter.js

@@ -30,6 +30,7 @@
 
 import { Squeeze } from '@hapi/good-squeeze';
 import { createWriteStream as writeStr } from 'fs';
+import { pipeline } from 'stream';
 
 import LogFormatJson from './log_format_json';
 import LogFormatString from './log_format_string';
@@ -51,18 +52,33 @@
   let dest;
   if (config.dest === 'stdout') {
     dest = process.stdout;
+    logInterceptor.pipe(squeeze).pipe(format).pipe(dest);
   } else {
     dest = writeStr(config.dest, {
       flags: 'a',
       encoding: 'utf8',
     });
 
-    logInterceptor.on('end', () => {
-      dest.end();
-    });
+    if (config.ignoreEnospcError) {
+      pipeline(logInterceptor, squeeze, format, dest, onFinished);
+    } else {
+      logInterceptor.on('end', () => {
+        dest.end();
+      });
+      logInterceptor.pipe(squeeze).pipe(format).pipe(dest);
+    }
   }
 
-  logInterceptor.pipe(squeeze).pipe(format).pipe(dest);
-
   return logInterceptor;
 }
+
+export function onFinished(error) {
+  if (error) {
+    if (error.code === 'ENOSPC') {
+      // eslint-disable-next-line no-console
+      console.error('Error in logging pipeline:', error.stack);
+    } else {
+      throw error;
+    }
+  }
+}

src/legacy/server/logging/log_reporter.test.js

@@ -0,0 +1,148 @@
+/*
+ * Copyright OpenSearch Contributors
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import os from 'os';
+import path from 'path';
+import fs from 'fs';
+import stripAnsi from 'strip-ansi';
+import { getLoggerStream, onFinished } from './log_reporter';
+
+const sleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms));
+
+describe('getLoggerStream', () => {
+  it('should log to stdout when the json config is set to false', async () => {
+    const lines = [];
+    const origWrite = process.stdout.write;
+    process.stdout.write = (buffer) => {
+      lines.push(stripAnsi(buffer.toString()).trim());
+      return true;
+    };
+
+    const loggerStream = getLoggerStream({
+      config: {
+        json: false,
+        dest: 'stdout',
+        filter: {},
+      },
+      events: { log: '*' },
+    });
+
+    loggerStream.end({ event: 'log', tags: ['foo'], data: 'test data' });
+
+    await sleep(500);
+
+    process.stdout.write = origWrite;
+    expect(lines.length).toBe(1);
+    expect(lines[0]).toMatch(/^log   \[[^\]]*\] \[foo\] test data$/);
+  });
+
+  it('should log to stdout when the json config is set to true', async () => {
+    const lines = [];
+    const origWrite = process.stdout.write;
+    process.stdout.write = (buffer) => {
+      lines.push(JSON.parse(buffer.toString().trim()));
+      return true;
+    };
+
+    const loggerStream = getLoggerStream({
+      config: {
+        json: true,
+        dest: 'stdout',
+        filter: {},
+      },
+      events: { log: '*' },
+    });
+
+    loggerStream.end({ event: 'log', tags: ['foo'], data: 'test data' });
+
+    await sleep(500);
+
+    process.stdout.write = origWrite;
+    expect(lines.length).toBe(1);
+    expect(lines[0]).toMatchObject({
+      type: 'log',
+      tags: ['foo'],
+      message: 'test data',
+    });
+  });
+
+  it('should log to custom file when the json config is set to false', async () => {
+    const dir = os.tmpdir();
+    const logfile = `dest-${Date.now()}.log`;
+    const dest = path.join(dir, logfile);
+
+    const loggerStream = getLoggerStream({
+      config: {
+        json: false,
+        dest,
+        filter: {},
+      },
+      events: { log: '*' },
+    });
+
+    loggerStream.end({ event: 'log', tags: ['foo'], data: 'test data' });
+
+    await sleep(500);
+
+    const lines = stripAnsi(fs.readFileSync(dest, { encoding: 'utf8' }))
+      .trim()
+      .split(os.EOL);
+    expect(lines.length).toBe(1);
+    expect(lines[0]).toMatch(/^log   \[[^\]]*\] \[foo\] test data$/);
+  });
+
+  it('should log to custom file when the json config is set to true and ignoreEnospcError', async () => {
+    const dir = os.tmpdir();
+    const logfile = `dest-${Date.now()}.log`;
+    const dest = path.join(dir, logfile);
+
+    const loggerStream = getLoggerStream({
+      config: {
+        json: true,
+        dest,
+        ignoreEnospcError: true,
+        filter: {},
+      },
+      events: { log: '*' },
+    });
+
+    loggerStream.end({ event: 'log', tags: ['foo'], data: 'test data' });
+
+    await sleep(500);
+
+    const lines = fs
+      .readFileSync(dest, { encoding: 'utf8' })
+      .trim()
+      .split(os.EOL)
+      .map((data) => JSON.parse(data));
+    expect(lines.length).toBe(1);
+    expect(lines[0]).toMatchObject({
+      type: 'log',
+      tags: ['foo'],
+      message: 'test data',
+    });
+  });
+
+  it('should handle ENOSPC error when disk full', () => {
+    const error = { code: 'ENOSPC', stack: 'Error stack trace' };
+    const consoleErrorSpy = jest.spyOn(console, 'error').mockImplementation();
+
+    expect(() => {
+      onFinished(error);
+    }).not.toThrow();
+
+    expect(consoleErrorSpy).toHaveBeenCalledWith('Error in logging pipeline:', 'Error stack trace');
+
+    consoleErrorSpy.mockRestore();
+  });
+
+  it('should throw error for non-ENOSPC error', () => {
+    const error = { message: 'non-ENOSPC error', code: 'OTHER', stack: 'Error stack trace' };
+
+    expect(() => {
+      onFinished(error);
+    }).toThrowError('non-ENOSPC error');
+  });
+});