Search in sources :

Example 1 with RetryPolicy

use of io.druid.indexing.common.RetryPolicy in project druid by druid-io.

the class KafkaIndexTaskClient method submitRequest.

private FullResponseHolder submitRequest(String id, HttpMethod method, String pathSuffix, String query, byte[] content, boolean retry) {
    final RetryPolicy retryPolicy = retryPolicyFactory.makeRetryPolicy();
    while (true) {
        FullResponseHolder response = null;
        Request request = null;
        TaskLocation location = TaskLocation.unknown();
        String path = String.format("%s/%s/%s", BASE_PATH, id, pathSuffix);
        Optional<TaskStatus> status = taskInfoProvider.getTaskStatus(id);
        if (!status.isPresent() || !status.get().isRunnable()) {
            throw new TaskNotRunnableException(String.format("Aborting request because task [%s] is not runnable", id));
        }
        try {
            location = taskInfoProvider.getTaskLocation(id);
            if (location.equals(TaskLocation.unknown())) {
                throw new NoTaskLocationException(String.format("No TaskLocation available for task [%s]", id));
            }
            // Netty throws some annoying exceptions if a connection can't be opened, which happens relatively frequently
            // for tasks that happen to still be starting up, so test the connection first to keep the logs clean.
            checkConnection(location.getHost(), location.getPort());
            try {
                URI serviceUri = new URI("http", null, location.getHost(), location.getPort(), path, query, null);
                request = new Request(method, serviceUri.toURL());
                // used to validate that we are talking to the correct worker
                request.addHeader(ChatHandlerResource.TASK_ID_HEADER, id);
                if (content.length > 0) {
                    request.setContent(MediaType.APPLICATION_JSON, content);
                }
                log.debug("HTTP %s: %s", method.getName(), serviceUri.toString());
                response = httpClient.go(request, new FullResponseHandler(Charsets.UTF_8), httpTimeout).get();
            } catch (Exception e) {
                Throwables.propagateIfInstanceOf(e.getCause(), IOException.class);
                Throwables.propagateIfInstanceOf(e.getCause(), ChannelException.class);
                throw Throwables.propagate(e);
            }
            int responseCode = response.getStatus().getCode();
            if (responseCode / 100 == 2) {
                return response;
            } else if (responseCode == 400) {
                // don't bother retrying if it's a bad request
                throw new IAE("Received 400 Bad Request with body: %s", response.getContent());
            } else {
                throw new IOException(String.format("Received status [%d]", responseCode));
            }
        } catch (IOException | ChannelException e) {
            // Since workers are free to move tasks around to different ports, there is a chance that a task may have been
            // moved but our view of its location has not been updated yet from ZK. To detect this case, we send a header
            // identifying our expected recipient in the request; if this doesn't correspond to the worker we messaged, the
            // worker will return an HTTP 404 with its ID in the response header. If we get a mismatching task ID, then
            // we will wait for a short period then retry the request indefinitely, expecting the task's location to
            // eventually be updated.
            final Duration delay;
            if (response != null && response.getStatus().equals(HttpResponseStatus.NOT_FOUND)) {
                String headerId = response.getResponse().headers().get(ChatHandlerResource.TASK_ID_HEADER);
                if (headerId != null && !headerId.equals(id)) {
                    log.warn("Expected worker to have taskId [%s] but has taskId [%s], will retry in [%d]s", id, headerId, TASK_MISMATCH_RETRY_DELAY_SECONDS);
                    delay = Duration.standardSeconds(TASK_MISMATCH_RETRY_DELAY_SECONDS);
                } else {
                    delay = retryPolicy.getAndIncrementRetryDelay();
                }
            } else {
                delay = retryPolicy.getAndIncrementRetryDelay();
            }
            String urlForLog = (request != null ? request.getUrl().toString() : String.format("http://%s:%d%s", location.getHost(), location.getPort(), path));
            if (!retry) {
                // if retry=false, we probably aren't too concerned if the operation doesn't succeed (i.e. the request was
                // for informational purposes only) so don't log a scary stack trace
                log.info("submitRequest failed for [%s], with message [%s]", urlForLog, e.getMessage());
                Throwables.propagate(e);
            } else if (delay == null) {
                log.warn(e, "Retries exhausted for [%s], last exception:", urlForLog);
                Throwables.propagate(e);
            } else {
                try {
                    final long sleepTime = delay.getMillis();
                    log.debug("Bad response HTTP [%s] from [%s]; will try again in [%s] (body/exception: [%s])", (response != null ? response.getStatus().getCode() : "no response"), urlForLog, new Duration(sleepTime).toString(), (response != null ? response.getContent() : e.getMessage()));
                    Thread.sleep(sleepTime);
                } catch (InterruptedException e2) {
                    Throwables.propagate(e2);
                }
            }
        } catch (NoTaskLocationException e) {
            log.info("No TaskLocation available for task [%s], this task may not have been assigned to a worker yet or " + "may have already completed", id);
            throw e;
        } catch (Exception e) {
            log.warn(e, "Exception while sending request");
            throw e;
        }
    }
}
Also used : Request(com.metamx.http.client.Request) Duration(org.joda.time.Duration) IOException(java.io.IOException) TaskStatus(io.druid.indexing.common.TaskStatus) IAE(io.druid.java.util.common.IAE) URI(java.net.URI) TaskLocation(io.druid.indexing.common.TaskLocation) ChannelException(org.jboss.netty.channel.ChannelException) IOException(java.io.IOException) FullResponseHandler(com.metamx.http.client.response.FullResponseHandler) FullResponseHolder(com.metamx.http.client.response.FullResponseHolder) RetryPolicy(io.druid.indexing.common.RetryPolicy) ChannelException(org.jboss.netty.channel.ChannelException)

Example 2 with RetryPolicy

use of io.druid.indexing.common.RetryPolicy in project druid by druid-io.

the class KafkaIndexTaskClient method pause.

public Map<Integer, Long> pause(final String id, final long timeout) {
    log.debug("Pause task[%s] timeout[%d]", id, timeout);
    try {
        final FullResponseHolder response = submitRequest(id, HttpMethod.POST, "pause", timeout > 0 ? String.format("timeout=%d", timeout) : null, true);
        if (response.getStatus().equals(HttpResponseStatus.OK)) {
            log.info("Task [%s] paused successfully", id);
            return jsonMapper.readValue(response.getContent(), new TypeReference<Map<Integer, Long>>() {
            });
        }
        final RetryPolicy retryPolicy = retryPolicyFactory.makeRetryPolicy();
        while (true) {
            if (getStatus(id) == KafkaIndexTask.Status.PAUSED) {
                return getCurrentOffsets(id, true);
            }
            final Duration delay = retryPolicy.getAndIncrementRetryDelay();
            if (delay == null) {
                log.error("Task [%s] failed to pause, aborting", id);
                throw new ISE("Task [%s] failed to pause, aborting", id);
            } else {
                final long sleepTime = delay.getMillis();
                log.info("Still waiting for task [%s] to pause; will try again in [%s]", id, new Duration(sleepTime).toString());
                Thread.sleep(sleepTime);
            }
        }
    } catch (NoTaskLocationException e) {
        log.error("Exception [%s] while pausing Task [%s]", e.getMessage(), id);
        return ImmutableMap.of();
    } catch (IOException | InterruptedException e) {
        log.error("Exception [%s] while pausing Task [%s]", e.getMessage(), id);
        throw Throwables.propagate(e);
    }
}
Also used : Duration(org.joda.time.Duration) ISE(io.druid.java.util.common.ISE) IOException(java.io.IOException) FullResponseHolder(com.metamx.http.client.response.FullResponseHolder) Map(java.util.Map) ImmutableMap(com.google.common.collect.ImmutableMap) RetryPolicy(io.druid.indexing.common.RetryPolicy)

Example 3 with RetryPolicy

use of io.druid.indexing.common.RetryPolicy in project druid by druid-io.

the class RemoteTaskActionClient method submit.

@Override
public <RetType> RetType submit(TaskAction<RetType> taskAction) throws IOException {
    log.info("Performing action for task[%s]: %s", task.getId(), taskAction);
    byte[] dataToSend = jsonMapper.writeValueAsBytes(new TaskActionHolder(task, taskAction));
    final RetryPolicy retryPolicy = retryPolicyFactory.makeRetryPolicy();
    while (true) {
        try {
            final Server server;
            final URI serviceUri;
            try {
                server = getServiceInstance();
                serviceUri = makeServiceUri(server);
            } catch (Exception e) {
                // Want to retry, so throw an IOException.
                throw new IOException("Failed to locate service uri", e);
            }
            final StatusResponseHolder response;
            log.info("Submitting action for task[%s] to overlord[%s]: %s", task.getId(), serviceUri, taskAction);
            try {
                response = httpClient.go(new Request(HttpMethod.POST, serviceUri.toURL()).setContent(MediaType.APPLICATION_JSON, dataToSend), new StatusResponseHandler(Charsets.UTF_8)).get();
            } catch (Exception e) {
                Throwables.propagateIfInstanceOf(e.getCause(), IOException.class);
                Throwables.propagateIfInstanceOf(e.getCause(), ChannelException.class);
                throw Throwables.propagate(e);
            }
            if (response.getStatus().getCode() / 100 == 2) {
                final Map<String, Object> responseDict = jsonMapper.readValue(response.getContent(), new TypeReference<Map<String, Object>>() {
                });
                return jsonMapper.convertValue(responseDict.get("result"), taskAction.getReturnTypeReference());
            } else {
                // Want to retry, so throw an IOException.
                throw new IOException(String.format("Scary HTTP status returned: %s. Check your overlord[%s] logs for exceptions.", response.getStatus(), server.getHost()));
            }
        } catch (IOException | ChannelException e) {
            log.warn(e, "Exception submitting action for task[%s]", task.getId());
            final Duration delay = retryPolicy.getAndIncrementRetryDelay();
            if (delay == null) {
                throw e;
            } else {
                try {
                    final long sleepTime = jitter(delay.getMillis());
                    log.info("Will try again in [%s].", new Duration(sleepTime).toString());
                    Thread.sleep(sleepTime);
                } catch (InterruptedException e2) {
                    throw Throwables.propagate(e2);
                }
            }
        }
    }
}
Also used : Server(io.druid.client.selector.Server) Request(com.metamx.http.client.Request) Duration(org.joda.time.Duration) IOException(java.io.IOException) URI(java.net.URI) ChannelException(org.jboss.netty.channel.ChannelException) URISyntaxException(java.net.URISyntaxException) IOException(java.io.IOException) StatusResponseHolder(com.metamx.http.client.response.StatusResponseHolder) StatusResponseHandler(com.metamx.http.client.response.StatusResponseHandler) RetryPolicy(io.druid.indexing.common.RetryPolicy) Map(java.util.Map) ChannelException(org.jboss.netty.channel.ChannelException)

Aggregations

RetryPolicy (io.druid.indexing.common.RetryPolicy)3 IOException (java.io.IOException)3 Duration (org.joda.time.Duration)3 Request (com.metamx.http.client.Request)2 FullResponseHolder (com.metamx.http.client.response.FullResponseHolder)2 URI (java.net.URI)2 Map (java.util.Map)2 ChannelException (org.jboss.netty.channel.ChannelException)2 ImmutableMap (com.google.common.collect.ImmutableMap)1 FullResponseHandler (com.metamx.http.client.response.FullResponseHandler)1 StatusResponseHandler (com.metamx.http.client.response.StatusResponseHandler)1 StatusResponseHolder (com.metamx.http.client.response.StatusResponseHolder)1 Server (io.druid.client.selector.Server)1 TaskLocation (io.druid.indexing.common.TaskLocation)1 TaskStatus (io.druid.indexing.common.TaskStatus)1 IAE (io.druid.java.util.common.IAE)1 ISE (io.druid.java.util.common.ISE)1 URISyntaxException (java.net.URISyntaxException)1