retry catalog queries and jdk downloads

reduce CI failures when Elastic JDK catalog is slow to respond by retrying.

Also retry download of the JDK itself (even though it doesn't typically fail).

Author: jsvd

build.gradle

@@ -47,6 +47,7 @@ import org.logstash.gradle.tooling.ExtractBundledJdkVersion
 import org.logstash.gradle.tooling.SignAliasDefinitions
 import org.logstash.gradle.tooling.ToolingUtils
 import org.logstash.gradle.tooling.SnapshotArtifactURLs
+import java.net.HttpURLConnection
 
 allprojects {
   group = 'org.logstash'
@@ -760,22 +761,88 @@ class JDKDetails {
         if (arch == "aarch64") {
             url += "-${arch}"
         }
-        println "Retrieving JDK from catalog..."
+        println "Retrieving JDK location URL from Elastic catalog..."
         def catalogMetadataUrl = URI.create(url).toURL()
-        def catalogConnection = catalogMetadataUrl.openConnection()
-        catalogConnection.requestMethod = 'GET'
-        if (catalogConnection.responseCode != 200) {
-            println "Can't find adoptiumjdk ${revision} for ${osName} on Elastic JVM catalog"
-            throw new GradleException("JVM not present on catalog")
+        int maxAttempts = 5
+        long retryDelayMillis = 3000
+        int connectTimeoutMillis = 15000
+        int readTimeoutMillis = 60000
+        int lastResponseCode = -1
+        String lastResponseMessage = null
+        String lastErrorBody = null
+        Exception lastException = null
+
+        for (int attempt = 1; attempt <= maxAttempts; attempt++) {
+            HttpURLConnection catalogConnection = null
+            try {
+                catalogConnection = catalogMetadataUrl.openConnection() as HttpURLConnection
+                catalogConnection.requestMethod = 'GET'
+                catalogConnection.connectTimeout = connectTimeoutMillis
+                catalogConnection.readTimeout = readTimeoutMillis
+
+                int responseCode = catalogConnection.responseCode
+                if (responseCode == 200) {
+                    def metadataRetrieved = catalogConnection.inputStream.withCloseable { stream -> stream.getText('UTF-8') }
+                    println "Retrieved!"
+
+                    def catalogMetadata = new JsonSlurper().parseText(metadataRetrieved)
+                    validateMetadata(catalogMetadata)
+
+                    return catalogMetadata.url
+                }
+
+                lastResponseCode = responseCode
+                lastResponseMessage = catalogConnection.responseMessage
+                lastErrorBody = catalogConnection.errorStream?.withCloseable { stream -> stream.getText('UTF-8') }?.trim()
+                lastException = null
+
+                def statusMessage = lastResponseMessage ? "${responseCode} ${lastResponseMessage}" : "${responseCode}"
+                if (lastErrorBody) {
+                    statusMessage += ": ${lastErrorBody.take(200)}"
+                }
+                if (attempt < maxAttempts) {
+                    println "Attempt ${attempt} of ${maxAttempts} failed with HTTP ${statusMessage}. Retrying..."
+                    try {
+                        Thread.sleep(retryDelayMillis)
+                    } catch (InterruptedException interruptedException) {
+                        Thread.currentThread().interrupt()
+                        throw new GradleException("Interrupted while retrying Elastic JVM catalog metadata request", interruptedException)
+                    }
+                }
+            } catch (Exception e) {
+                lastException = e
+                lastResponseCode = -1
+                lastResponseMessage = null
+                lastErrorBody = null
+                if (attempt < maxAttempts) {
+                    println "Attempt ${attempt} of ${maxAttempts} failed: ${e.message}. Retrying..."
+                    try {
+                        Thread.sleep(retryDelayMillis)
+                    } catch (InterruptedException interruptedException) {
+                        Thread.currentThread().interrupt()
+                        throw new GradleException("Interrupted while retrying Elastic JVM catalog metadata request", interruptedException)
+                    }
+                }
+            } finally {
+                catalogConnection?.disconnect()
+            }
         }
 
-        def metadataRetrieved = catalogConnection.content.text
-        println "Retrieved!"
-
-        def catalogMetadata = new JsonSlurper().parseText(metadataRetrieved)
-        validateMetadata(catalogMetadata)
-
-        return catalogMetadata.url
+        def requestedArtifact = "adoptiumjdk-${revision}+${build}-${osName}${arch == "aarch64" ? "-${arch}" : ""}"
+        if (lastResponseCode == 404) {
+            throw new GradleException("Elastic JVM catalog returned 404 for ${requestedArtifact}. Verify the artifact exists.", lastException)
+        }
+        if (lastResponseCode != -1) {
+            def statusMessage = lastResponseMessage ? "${lastResponseCode} ${lastResponseMessage}" : "${lastResponseCode}"
+            if (lastErrorBody) {
+                statusMessage += " (${lastErrorBody.take(200)})"
+            }
+            throw new GradleException("Failed to fetch Elastic JVM catalog metadata for ${requestedArtifact}. Last HTTP response: ${statusMessage}", lastException)
+        }
+        if (lastException != null) {
+            throw new GradleException("Failed to fetch Elastic JVM catalog metadata for ${requestedArtifact}: ${lastException.message}", lastException)
+        }
+        throw new GradleException("Failed to fetch Elastic JVM catalog metadata for ${requestedArtifact}.")
     }
 
     //Verify that the artifact metadata correspond to the request, if not throws an error
@@ -848,6 +915,7 @@ tasks.register("downloadJdk", Download) {
     src project.ext.jdkURL
     onlyIfNewer true
     overwrite false
+    retries 3
     quiet true
     inputs.file("${projectDir}/versions.yml")
     outputs.file(project.ext.jdkDownloadLocation)