add hacky workaround for EOF errors in kubectl cp

Azure · Oct 27, 2023 · ca4aac7 · ca4aac7
1 parent 4835135
commit ca4aac7
Show file tree

Hide file tree

Showing 3 changed files with 36 additions and 4 deletions.
diff --git a/src/commands/aksTCPCollection/tcpDumpCollection.ts b/src/commands/aksTCPCollection/tcpDumpCollection.ts
@@ -6,7 +6,7 @@ import { getExtension } from '../utils/host';
 import { Errorable, failed, map as errmap } from '../utils/errorable';
 import * as tmpfile from '../utils/tempfile';
 import { TcpDumpDataProvider, TcpDumpPanel } from '../../panels/TcpDumpPanel';
-import { invokeKubectlCommand } from '../utils/kubectl';
+import { getVersion, invokeKubectlCommand } from '../utils/kubectl';
 
 export async function aksTCPDump(_context: IActionContext, target: any) {
     const kubectl = await k8s.extension.kubectl.v1;
@@ -47,7 +47,13 @@ export async function aksTCPDump(_context: IActionContext, target: any) {
         return;
     }
 
-    const dataProvider = new TcpDumpDataProvider(kubectl, kubeConfigFile.filePath, clusterInfo.result.name, linuxNodesList.result);
+    const kubectlVersion = await getVersion(kubectl, kubeConfigFile.filePath);
+    if (failed(kubectlVersion)) {
+        vscode.window.showErrorMessage(kubectlVersion.error);
+        return;
+    }
+
+    const dataProvider = new TcpDumpDataProvider(kubectl, kubeConfigFile.filePath, kubectlVersion.result, clusterInfo.result.name, linuxNodesList.result);
     const panel = new TcpDumpPanel(extension.result.extensionUri);
 
     panel.show(dataProvider, kubeConfigFile);

diff --git a/src/commands/utils/kubectl.ts b/src/commands/utils/kubectl.ts
@@ -14,6 +14,22 @@ type KubeconfigCommandConfig = {
     exitCodeBehaviour: NonZeroExitCodeBehaviour
 };
 
+export type K8sVersion = {
+    major: string,
+    minor: string,
+    gitVersion: string,
+    buildDate: string
+};
+
+export type KubectlVersion = {
+    clientVersion: K8sVersion,
+    serverVersion: K8sVersion
+};
+
+export function getVersion(kubectl: APIAvailable<KubectlV1>, kubeConfigFile: string): Promise<Errorable<KubectlVersion>> {
+    return getKubectlJsonResult(kubectl, kubeConfigFile, "version -o json");
+}
+
 export async function getExecOutput(kubectl: APIAvailable<KubectlV1>, kubeConfigFile: string, namespace: string, pod: string, podCommand: string): Promise<Errorable<KubectlV1.ShellResult>> {
     const plainCommand = `exec -n ${namespace} ${pod} -- ${podCommand}`;
     const config: KubeconfigCommandConfig = {

diff --git a/src/panels/TcpDumpPanel.ts b/src/panels/TcpDumpPanel.ts
@@ -2,10 +2,11 @@ import { platform } from "os";
 import { relative } from "path";
 import { Uri, commands, window, workspace } from "vscode";
 import * as k8s from 'vscode-kubernetes-tools-api';
+import * as semver from 'semver';
 import { failed, map as errmap, Errorable } from "../commands/utils/errorable";
 import { MessageHandler, MessageSink } from "../webview-contract/messaging";
 import { BasePanel, PanelDataProvider } from "./BasePanel";
-import { getExecOutput, invokeKubectlCommand } from "../commands/utils/kubectl";
+import { KubectlVersion, getExecOutput, invokeKubectlCommand } from "../commands/utils/kubectl";
 import { CompletedCapture, InitialState, ToVsCodeMsgDef, ToWebViewMsgDef } from "../webview-contract/webviewDefinitions/tcpDump";
 import { withOptionalTempFile } from "../commands/utils/tempfile";
 
@@ -54,6 +55,7 @@ export class TcpDumpDataProvider implements PanelDataProvider<"tcpDump"> {
     constructor(
         readonly kubectl: k8s.APIAvailable<k8s.KubectlV1>,
         readonly kubeConfigFilePath: string,
+        readonly kubectlVersion: KubectlVersion,
         readonly clusterName: string,
         readonly linuxNodesList: string[]
     ) { }
@@ -350,7 +352,15 @@ spec:
         }
 
         const localCpPath = getLocalKubectlCpPath(localCaptureUri);
-        const command = `cp -n ${debugPodNamespace} ${getPodName(node)}:${captureFileBasePath}${captureName}.cap ${localCpPath}`;
+
+        // `kubectl cp` can fail with an EOF error for large files, and there's currently no good workaround:
+        // See: https://github.com/kubernetes/kubernetes/issues/60140
+        // The best advice I can see is to use the 'retries' option if it is supported, and the
+        // 'request-timeout' option otherwise.
+        const clientVersion = this.kubectlVersion.clientVersion.gitVersion.replace(/^v/, "");
+        const isRetriesOptionSupported = semver.parse(clientVersion) && semver.gte(clientVersion, "1.23.0");
+        const cpEOFAvoidanceFlag = isRetriesOptionSupported ? "--retries 99" : "--request-timeout=10m";
+        const command = `cp -n ${debugPodNamespace} ${getPodName(node)}:${captureFileBasePath}${captureName}.cap ${localCpPath} ${cpEOFAvoidanceFlag}`;
         const output = await invokeKubectlCommand(this.kubectl, this.kubeConfigFilePath, command);
         if (failed(output)) {
             webview.postDownloadCaptureFileResponse({