Streaming partial results

A long-running task can publish intermediate results as it works; a consumer polls them as they arrive — useful for ETL, ML training steps, or batch progress. In Java, partials ride on task logs: durable, per-job log rows in shared storage.

Publish

From inside the handler, write a log row against the job (get the id across with the middleware pattern):

for (int epoch = 1; epoch <= epochs; epoch++) {
    double loss = trainEpoch(epoch);
    flexiq.writeTaskLog(jobId, "train", "result",
            "{\"epoch\": " + epoch + ", \"loss\": " + loss + "}");
}

writeTaskLog(jobId, taskName, level, message) takes any level string; an overload adds an extra payload. By the cross-SDK contract, published partials use level "result" — workers in other processes that publish partials store them the same way, so they are readable here.

Consume

getTaskLogs(jobId) returns every row for the job, oldest first; poll it until the job is terminal:

String id = flexiq.enqueue(train, config);

int seen = 0;
while (flexiq.getJob(id).map(j -> j.status == JobStatus.PENDING || j.status == JobStatus.RUNNING).orElse(false)) {
    List<TaskLog> logs = flexiq.getTaskLogs(id);
    for (TaskLog log : logs.subList(seen, logs.size())) {
        if ("result".equals(log.level)) {
            render(log.message);
        }
    }
    seen = logs.size();
    Thread.sleep(200);
}

int result = flexiq.getResult(id, Integer.class).orElseThrow();

It works from any process sharing the storage, and a late consumer still reads every value already published. Each TaskLog carries jobId, taskName, level, message, extra, and loggedAt (Unix ms).