We try to decrease the network usage with avoiding sending out
to much tiles to the client. When we already sent out two versions
of the same tile without having the tileprocessed message from the
client we delay sending out the next version to avoid spamming tiles
on the network.
Change-Id: Ia47cd7c0d3fb829f6777f0c3265970433591df19
It's good if this limit big enough to send all the tiles of the
current visible area at once, so the tiles arrive at the same
time to the client by zoom or scroll. Now this value is set to a bit
bigger so if we have a small amount of tiles before zoom / scroll we'll
still see the same on the client side.
Change-Id: I8e28dbf6197fe2f683fe9528e9a32c15a191b20e
I changed the code in this commit:
c2a5f6acb0
To make kit send a tilecombine message even if it does not
send actual tile data so we can track that the rendering was
done and so we can update the clients' _tilesBeingRendered
list. The issue is that tileBeingRendered object
belongs to not only one client, but more and so we don't
know which client gets the actual empty tile response.
So revert this method and rather use a smaller timeout for "waiting" the
arrival of the rendered tile.
Change-Id: I2dbbab1a62b81cbbb5314f2f37fdbc3415c69130
When we're doing the prerendering we also need to register
a tileBeingRendered object, so we know that the given tile
will arrive soon. In earlier version of the code, rendering
and sending of the tile was not separated in time, but now it is,
so we need create a tileBeingRendered object even if we don't
subscribe the client to it yet.
Change-Id: I1374c22e5ffebe1d6fcc6e37261ddcedeb9066ec
In this case the part number is undefined, somewhere it is set to 0 or 1,
but has no meaning at all. Just avoid using it with text ducments.
Change-Id: Ic98217bf3ea6c86d37c34e42302bf456f7274975
Re-think the plumbing between the different parts of the C++ Online
code. Do try to have it work more like in real Online on all but the
lowest socket level. Except that we don't have multiple processes, but
threads inside the same process. And instead of using actual system
sockets for WebSocket traffic between the threads, we use our own
FakeSocket things, with no WebSocket framing of messages.
Reduce the amount of #ifdef MOBILEAPP a bit also by compiling in the
UnitFoo things. Hardcode that so that no unit testing is ever
attempted, though. We don't try to dlopen any library.
Corresponding changes in the app Objective-C code. Plus fixes and
functionality improvements.
Now it gets so far that the JavaScript code thinks it has the document
tiles presented, and doesn't crash. But it hangs occasionally. And all
tiles show up blank.
Anyway, progress.
Change-Id: I769497c9a46ddb74984bc7af36d132b7b43895d4
Re-think Linux vs mobile ifdefs a bit. Use #ifdef __linux only to
surround code that actually is Linux-specific. Use #ifdef MOBILEAPP
for code that is for a mobile version (with no separste wsd, forkit,
and kit processes, and with no WebSocket protocol used).
Bypass UnitFoo for mobile. Possibly we do want the UnitFoo stuff after
all on mobile, to run in some special testing mode? Hard to say, let's
skipt it for now.
So we can avoid that tile sending stop working because server is
waiting for tileprocessed messages which will not arrive.
Change-Id: I545346c50d49340999608aadac32b5190ede43c5
We don't check if boost is available in the configure phase so we
can't use it. It is possible to add it but as we only use it in
one place and it is actually not really needed, it is better to
remove the usage.
This change also moves the std::list to std::deque.
Change-Id: I4c661a2fb8995bf756a0815a6e30bb304f22d3c8
Reviewed-on: https://gerrit.libreoffice.org/58640
Reviewed-by: Ashod Nakashian <ashnakash@gmail.com>
Tested-by: Ashod Nakashian <ashnakash@gmail.com>
This reverts commit f1a385be98.
It's not easy, it can interfere the tile tracking.
Change-Id: I1e4ec9b4d66e5e912873f673fd5cb71ba55a9332
Reviewed-on: https://gerrit.libreoffice.org/59326
Reviewed-by: Tamás Zolnai <tamas.zolnai@collabora.com>
Tested-by: Tamás Zolnai <tamas.zolnai@collabora.com>
Store the tile cache names and drop outdated tiles
times to times, so we can avoid tile rendering / sending to
stuck.
Change-Id: Ibff001307c7c660cbc57ab20c29c430e0090444d
SenderQueue might drop some tiles, so we were waiting for
tileprocessed message for a tile which was not sent at all.
Change-Id: I7c502966f656e46df7c22002dee19aeabbf97774
I used number 25 as this limit. It's an approximate value. It's enough
to handle the first 5-10 character input without waiting for tileprocessed
messages. After the first 5-10 characters the tileprocessed messages are
arriving continuously (same frequency as the typing speed), so for the later
character inputs we always have some tileprocessed messages arrived so
we can push new tiles to the network again.
This 25 upper limit also seems enough to send all tiles when a whole page is
requested (e.g. zoom os scroll).
We store the requested tiles in a list, used as a queue. I don't use std::queue
because sometimes we need to do deduplication (replace older versions of the
same tile with a newer version).
Change-Id: I22ff3e35c8ded6001c7fc160abdc1f1b12ce3bae
We always one bunch of tiles (e.g. all tiles invalidated) and we
are waiting until client send tileprocessed message back for all
tiles before sending the new tiles.
By canceltiles message we drop every previously requested tiles and
make wsd ready to send new tiles, which will be requested by the client
in theory.
Change-Id: I9901420ada549e962ffaf5e6bd58e52b86bd129d
Since changing the tracking state is done by toggling,
we need to wait to get the current state at load time
before we can tell whether we need to toggle it or not.
Change-Id: Ib5a2639b2acf3874c191971eedf9a3bebcefebad
Reviewed-on: https://gerrit.libreoffice.org/53415
Reviewed-by: Jan Holesovsky <kendy@collabora.com>
Tested-by: Jan Holesovsky <kendy@collabora.com>
The bool flag was causing 2 complete separate code paths anyway.
Also remove stop(), calling stop() followed by close() made no
difference.
Change-Id: Ica4c887b0324390d4e006a26eb4119bd5ab08723
Just broadcast the message and let clients deal with it. This is similar
to how we do it when we find the document conflict via CheckFileInfo
Change-Id: I52855fcb96a359b3915afe71d481321f79b4554b
... instead of directly stop()ing it. The close request approach also
makes sure that outgoing document is saved to storage.
Change-Id: I44f61db00dbd326dec80f59f4a2cbb617048aa94
Reviewed-on: https://gerrit.libreoffice.org/49122
Reviewed-by: Ashod Nakashian <ashnakash@gmail.com>
Tested-by: Ashod Nakashian <ashnakash@gmail.com>
This also prevents saving the document back to storage.
Without this patch, in case of document conflict (when the document is
changed behind our back in the storage), the unmodified document end up
being saved to storage, hence overwriting the document which was changed
behind our back.
Change-Id: Ie027f5b49c37e7df3f36499e0fef3eca78173969
Reviewed-on: https://gerrit.libreoffice.org/49114
Reviewed-by: Ashod Nakashian <ashnakash@gmail.com>
Tested-by: Ashod Nakashian <ashnakash@gmail.com>
* logs helpful messages for various error corner-cases.
* optimized file descriptor closing for large fd counts.
Change-Id: I8cba9ecb3d71ddc6e22e20d89368d8c6b9b5097f
Now all logging is done after checking if the
level in question is enabled or not (thanks to
the macros LOG_XXX), which saves unnecessary
conversions and stringification when said level
is disabled.
Change-Id: Icde31e067f60269563896f04f8b0d65643720766
Reviewed-on: https://gerrit.libreoffice.org/47885
Reviewed-by: Michael Meeks <michael.meeks@collabora.com>
Tested-by: Michael Meeks <michael.meeks@collabora.com>
We don't force saving unconditionally now. Only
when the doc is reasonably expected to be modified
do we force saving (to circumvent the minimum duration
between auto-saves).
We invoke auto-saving before stopping the DocBroker
polling loop, whether due to idleness or server recycling.
Change-Id: I257d55f190d3df6a3ba82f2666c7602da0581d0c
Reviewed-on: https://gerrit.libreoffice.org/47887
Reviewed-by: Michael Meeks <michael.meeks@collabora.com>
Tested-by: Michael Meeks <michael.meeks@collabora.com>
Now always given a proper reason too.
Also, stop polling thread and cleanup when
failing to acquire/spawn a child process.
Change-Id: I7ddee01dd47b8ee72f2d9134c0f1b264634d8611
Reviewed-on: https://gerrit.libreoffice.org/47886
Reviewed-by: Michael Meeks <michael.meeks@collabora.com>
Tested-by: Michael Meeks <michael.meeks@collabora.com>
Previously we assumed we are saving based on
lastSaveTime, which is incorrect because it
is set only upon successful saving and storing,
which might fail.
Now lastSaveResponseTime is used to track whether
there are saving requests in flight. And lastSaveTime
is only used when we do store the document in storage.
Change-Id: I73e5c04432981d0cca11b8cf854414738bd894de
Reviewed-on: https://gerrit.libreoffice.org/47884
Reviewed-by: Michael Meeks <michael.meeks@collabora.com>
Tested-by: Michael Meeks <michael.meeks@collabora.com>
Maintaining a flag is problematic, at least the way
it was reset on adding new sessions. Luckily there
should be no reason for having it, since we should
check it only when removing sessions. Also, we
need to check for every case of removal, and not, as
was, just on disconnection, which is incomplete.
Change-Id: I878766701228c41fc93eeaff21852fa887de9eff
Reviewed-on: https://gerrit.libreoffice.org/47883
Reviewed-by: Michael Meeks <michael.meeks@collabora.com>
Tested-by: Michael Meeks <michael.meeks@collabora.com>