I narrowed down my problem to the async fetcher class which is derived from
AsyncReceiver class.
In my plugin, I maintain a few maps which get periodically updated. In my
plugin, I create a thread
to constantly (every 30s) poll an external server. When the thread wakes
up, it uses the async fetcher
class to retrieve info from an external server to update my maps. The
following two lines seem
to be causing the issue:
AsyncHttpFetch* fetcher = new AsyncHttpFetch(mUrl,
AsyncHttpFetch::STREAMING_DISABLED);
Async::execute<AsyncHttpFetch>(this, fetcher, mMutex);
If I comment them out, there's no leak. My async fetcher class follows a
similar pattern as some
of the atscppapi examples. To narrow down the issue, I've created a small
plugin which models
my plugin and also creates the issue I am seeing in my plugin. I've
attached my test-plugin files.
I would appreciate it if someone can take a look at my implementation and
let me know I am
using the API incorrectly.
Thank you.
On Tue, Mar 5, 2019 at 10:52 AM Dk Jack <dnj0...@gmail.com> wrote:
> Thanks Leif,
> I'd like to understand a bit more before introducing another variable. The
> thing that puzzles me the most is why the system objects are behaving
> erratically when I am not modifying them or holding on to them..
>
> On Mon, Mar 4, 2019 at 7:11 PM Leif Hedstrom <zw...@apache.org> wrote:
>
>> Before going down this rathole, I’d give v7.1.6 or v8.0.2 a spin.
>> Remember, v6.x is an unsupported version, there is no more development nor
>> any more releases.
>>
>> Cheers,
>>
>> — Leif
>>
>> > On Mar 4, 2019, at 20:00, Dk Jack <dnj0...@gmail.com> wrote:
>> >
>> > I am using atscppapi's for my plugin. I do something like this...
>> >
>> > Headers& headers = transaction.getClientRequest().getHeaders();
>> >
>> >
>> >
>> > On Mon, Mar 4, 2019 at 6:09 PM Pushkar Pradhan
>> > <pprad...@verizonmedia.com.invalid> wrote:
>> >
>> >> I am also trying to learn the TS APIs. I think you are supposed to
>> call the
>> >> TS...Release if you get a reference.
>> >> E.g.
>> >> if (TS_SUCCESS == TSHttpTxnClientReqGet(txnp, &buffer, &hdr_loc)) {
>> >> TSDebug(PLUGIN_NAME, "Found client request");
>> >> txn_info += R"(,"client-request":)" + collect_headers(buffer,
>> >> hdr_loc, TSHttpTxnClientReqBodyBytesGet(txnp));
>> >> TSHandleMLocRelease(buffer, TS_NULL_MLOC, hdr_loc);
>> >> buffer = nullptr;
>> >> }
>> >>
>> >>
>> >>> On Mon, Mar 4, 2019 at 6:06 PM Dk Jack <dnj0...@gmail.com> wrote:
>> >>>
>> >>> Thanks Pushkar, I'll try that...
>> >>>
>> >>> On Mon, Mar 4, 2019 at 4:44 PM Pushkar Pradhan
>> >>> <pprad...@verizonmedia.com.invalid> wrote:
>> >>>
>> >>>> Perhaps compiling with LSAN will help: --enable-lsan
>> enable
>> >>>> stand-alone Leak Sanitizer
>> >>>>
>> >>>>
>> >>>>> On Mon, Mar 4, 2019 at 4:42 PM Bryan Call <bc...@apache.org> wrote:
>> >>>>>
>> >>>>> Looking at the graphs it looks like your headers and http1client
>> >>> sessions
>> >>>>> are leaking. Have you looked at this example to on how to send back
>> >>>> error
>> >>>>> messages using the C++ API:
>> >>>>>
>> >>>>
>> >>>
>> >>
>> https://github.com/apache/trafficserver/blob/master/example/cppapi/customresponse/CustomResponse.cc
>> >>>>>
>> >>>>> -Bryan
>> >>>>>
>> >>>>>> On Mar 4, 2019, at 3:50 PM, Dk Jack <dnj0...@gmail.com> wrote:
>> >>>>>>
>> >>>>>> Hi,
>> >>>>>> I am trying to chase a memory leak issue. I am using ATS version
>> >>> 6.2.3
>> >>>>> and
>> >>>>>> I have a custom plugin written using the c++ API (atscppapi)
>> >> running
>> >>> on
>> >>>>> the
>> >>>>>> same server. I enabled memory dump in my configuration and sent
>> >>> traffic
>> >>>>>> into the ATS with and without my plugin for an hour. Later, I
>> >>> extracted
>> >>>>> the
>> >>>>>> individual memory stats and plotted the stats for each object type
>> >>>>> against
>> >>>>>> time. The link shown below contains side-by-side plots for each ATS
>> >>>>> object
>> >>>>>> type without and with my plugin (the right side figures are with my
>> >>>>> plugin
>> >>>>>> enabled).
>> >>>>>>
>> >>>>>>
>> >>>
>> https://www.dropbox.com/s/rfszd6fnjydm8s7/ats_mem_stats_comparison.pdf
>> >>>>>>
>> >>>>>> Without my plugin, the memory consumption increases for various
>> >>> objects
>> >>>>> in
>> >>>>>> the beginning and then tapers off. However, with my plugin enabled,
>> >>> the
>> >>>>>> memory consumption is higher and more erratic. I am not holding on
>> >> to
>> >>>> the
>> >>>>>> request in my plugin. My plugin attaches TransactionPlugin to some
>> >>>>>> transactions (sending back 4xx response to client). For most
>> >> others,
>> >>> it
>> >>>>>> lets it pass through. My plugin, doesn't allocate much memory. The
>> >>>> memory
>> >>>>>> it allocates is sort of semi-static (manages a few maps that are
>> >>>> updated
>> >>>>>> periodically). I don't anticipate my plugin's memory consumption to
>> >>>>>> increase with the transactions.
>> >>>>>>
>> >>>>>> Since the presence of my plugin is the affecting the memory stats
>> >> of
>> >>>>>> internal ATS objects, I am suspecting this could result of me not
>> >>> using
>> >>>>> the
>> >>>>>> API correctly. Would appreciate if anyone can venture a guess as to
>> >>>> what
>> >>>>>> could be happening. Thank for help.
>> >>>>>>
>> >>>>>> Dk.
>> >>>>>
>> >>>>>
>> >>>>
>> >>>
>> >>
>>
>>
