On Mon, Feb 22, 2016 at 11:53 AM, Colin P. McCabe <[email protected]> wrote:
> On Mon, Feb 22, 2016 at 9:33 AM, Sean Busbey <[email protected]> wrote: > > Is the expected overhead something like this: > > > > # rolling upgrade of system with configs in place that disable tracing > > # upgrade tracing service > > # rolling restart of system with configs in place to re-enable tracing > > > > An extra rolling restart just due to tracing does seem like a large > request. > > There is no extra rolling restart. I would expect the workflow to be > something like: > 1. bring down old htraced > 2. rolling upgrade of HDFS system > 3. bring up new htraced > > Colin > Ah. so we rely on the client libs degrading gracefully when the server is away to avoid the restart. That makes sense. -- Sean
