我用Java编写了一个servlet代码,用于从存储在Google Cloud Storage中的文件中读取一行。一旦我读取了每一行,我就把它传递给预测API。一旦我得到预测的文本通过。我将它附加到原始行,并将其存储在Google云存储中的其他文件中。
此源文件为csv格式,包含超过10,000条记录。由于我单独解析它,将其传递给预测API,然后存储回云存储。这样做要花很多时间。由于App Engine有30节的限制,任务队列也有限制。谁能给我一些建议吗?因为重新启动程序不是一个选项,因为我无法从我停止的地方启动预测。
下面是我的代码:
@SuppressWarnings("serial")
public class PredictionWebAppServlet extends HttpServlet {
private static final String APPLICATION_NAME = "span-test-app";
static final String MODEL_ID = "span-senti";
static final String STORAGE_DATA_LOCATION = "/bigdata/training_set/";
private static HttpTransport httpTransport;
private static final JsonFactory JSON_FACTORY = JacksonFactory
.getDefaultInstance();
public static final String INPUT_BUCKETNAME = "bigdata";
public static final String INPUT_FILENAME = "abc.csv";
public static final String OUTPUT_BUCKETNAME = "bigdata";
public static final String OUTPUT_FILENAME = "def.csv";
private static Credential authorize() throws Exception {
Credential cr = new GoogleCredential.Builder()
.setTransport(httpTransport)
.setJsonFactory(JSON_FACTORY)
.setServiceAccountId(
"878482284233-aacp8vd5297aqak7v5r0f507qr63mab4@developer.gserviceaccount.com")
.setServiceAccountScopes(
Collections.singleton(PredictionScopes.PREDICTION))
.setServiceAccountPrivateKeyFromP12File(
new File(
"28617ba6faac0a51eb2208edba85d2e20e6081b4-privatekey.p12"))
.build();
return cr;
}
public void doGet(HttpServletRequest req, HttpServletResponse resp)
throws IOException {
try {
httpTransport = GoogleNetHttpTransport.newTrustedTransport();
Credential credential = authorize();
Prediction prediction = new Prediction.Builder(httpTransport,
JSON_FACTORY, credential).setApplicationName(APPLICATION_NAME)
.build();
GcsService gcsService = GcsServiceFactory.createGcsService();
GcsFilename filename = new GcsFilename(INPUT_BUCKETNAME, INPUT_FILENAME);
GcsFilename filename1 = new GcsFilename(OUTPUT_BUCKETNAME,
OUTPUT_FILENAME);
GcsFileOptions options = new GcsFileOptions.Builder()
.mimeType("text/html").acl("public-read")
.addUserMetadata("myfield1", "my field value").build();
GcsOutputChannel writeChannel = gcsService.createOrReplace(filename1, options);
PrintWriter writer = new PrintWriter(Channels.newWriter(writeChannel,
"UTF8"));
GcsInputChannel readChannel = null;
BufferedReader reader = null;
readChannel = gcsService.openReadChannel(filename, 0);
reader = new BufferedReader(Channels.newReader(readChannel, "UTF8"));
String line;
String cvsSplitBy = ",";
String temp_record = "";
Input input = new Input();
InputInput inputInput = new InputInput();
while ((line = reader.readLine()) != null) {
String[] post = line.split(cvsSplitBy);
inputInput.setCsvInstance(Collections
.<Object> singletonList(post[1]));
input.setInput(inputInput);
Output output = prediction.trainedmodels()
.predict("878482284233", MODEL_ID, input).execute();
for (int i = 0; i < 10; i++) {
temp_record = temp_record + post[i] + ",";
}
temp_record = temp_record + output.getOutputLabel();
writer.println(temp_record);
}
writer.flush();
writer.close();
//resp.getWriter().println(temp_record);
} catch (Exception e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
finally{
}
}
}
你自己也在暗示。
如果你认为你的工作可以在10分钟内完成,你可以单独使用任务队列。
如果没有,您将需要使用任务队列和后端组合。您需要将其推送到后端实例中。看看Push队列和后端
UPDATE -使用模块代替后端
后端不推荐使用模块。在模块中这样做的一种方法是:
- 将你的应用程序转换为模块结构
- 定义一个手动缩放的模块
- 处理该模块中的"/_ah/start" url
- 执行"/_ah/start"处理程序中的所有作业
手动伸缩实例对它们可能运行的时间没有限制。如果实例具有手动缩放,则可以在"/_ah/start"请求中运行"永远"。嘿,你甚至可以开始线程,如果你喜欢的话。但这对这份工作来说不应该是必需的。
这正是MapReduce框架的作用。