These saved documents are then removed to free some memory when next
memory cleanup is triggered in the admin loop.
Change-Id: Ia789a65dc204b546f1d3627a8006c4590bdca371
Start killing documents when memory usage goes above threshold.
Also make it possible to close documents from admin instance.
In DocumentBroker::closeDocument, just set the _stop flag and wake
up the polling thread which will terminate the children, instead of
manually terminating the children.
Change-Id: Ie70e05b3fb6ea816a87b6dcfaed92cdddb94aa90
I guess checking for "{" in parseJSON() happens for performance reasons,
but then it's a good idea to also avoid constructing a temporary string
for no reason.
Change-Id: I285ba0ed615054fc014796f7d8971a2b9783b075
With help from Valgrind to find and verify
these leaks. Also some minor cleanup to the
deflate logic.
Change-Id: I3bb3f1e01cef7025c45874ce52cfc922dfd19e21
Reviewed-on: https://gerrit.libreoffice.org/39465
Reviewed-by: Ashod Nakashian <ashnakash@gmail.com>
Tested-by: Ashod Nakashian <ashnakash@gmail.com>
With help from Valgrind to find and verify
these leaks.
Change-Id: I3afeed89dc4bcd714a222f81822144477a346fb0
Reviewed-on: https://gerrit.libreoffice.org/39464
Reviewed-by: Ashod Nakashian <ashnakash@gmail.com>
Tested-by: Ashod Nakashian <ashnakash@gmail.com>
If Forkit is dead, getNewChild will fail fast.
Try not to busy loop in that case and yield
the CPU for a nominal duration, which currently
is the tenth of the normal getNewChild timeout.
Change-Id: I1a94dfedbf2a4f4fc12e4d33d1307f70c307987a
Reviewed-on: https://gerrit.libreoffice.org/39248
Reviewed-by: Ashod Nakashian <ashnakash@gmail.com>
Tested-by: Ashod Nakashian <ashnakash@gmail.com>
This is called later after the while loop. Breaking out of while loop
should be enough.
Change-Id: I04979d3af1f475c05b5a43d7afe47770ff69ee25
Reviewed-on: https://gerrit.libreoffice.org/39086
Reviewed-by: pranavk <pranavk@collabora.co.uk>
Tested-by: pranavk <pranavk@collabora.co.uk>
Changes protocol to use 'wid' instead of 'hash' everywhere. Wire-ids
are monotonically increasing integers that can be mapped to hash
values for all of the hash values and tiles we cache internally.
Change-Id: Ibcb25817bab0f453e93d52a6f99d3ff65059e47d
Also, no need for the DocBrokerMutex to be locked
in createForkit.
Change-Id: I651c50c8b5518f7b0ee0d8b15c413fc070e30443
Reviewed-on: https://gerrit.libreoffice.org/38993
Reviewed-by: Ashod Nakashian <ashnakash@gmail.com>
Tested-by: Ashod Nakashian <ashnakash@gmail.com>
LOOLWSD::checkAndRestoreForKit() eventually takes
NewChildrenLock, which is also taken in getNewChild_Blocks().
Luckily, we can defer that to the Prisoner poll, which we do.
Change-Id: Ib65c0be08870d1c4c68efd72192dac2e73c79f7c
Reviewed-on: https://gerrit.libreoffice.org/38992
Reviewed-by: Ashod Nakashian <ashnakash@gmail.com>
Tested-by: Ashod Nakashian <ashnakash@gmail.com>
Valgrind spotted one case, and the other is possible but
not common it seems.
Change-Id: Id5e41145f597c3564263adb25b7b765db1c90bf7
Reviewed-on: https://gerrit.libreoffice.org/38991
Reviewed-by: Ashod Nakashian <ashnakash@gmail.com>
Tested-by: Ashod Nakashian <ashnakash@gmail.com>
To make the debugging easier...
Change-Id: I7c6e748e72a595a6c3a5942a20874339b8456f19
Reviewed-on: https://gerrit.libreoffice.org/38781
Reviewed-by: Jan Holesovsky <kendy@collabora.com>
Tested-by: Jan Holesovsky <kendy@collabora.com>
Seems to have no effect, so gone in
favor of RLIMIT_AS (virtual memeory).
Change-Id: I210879ec9285f420c9f9839cdabf45c42d865fb3
Reviewed-on: https://gerrit.libreoffice.org/38720
Reviewed-by: Ashod Nakashian <ashnakash@gmail.com>
Tested-by: Ashod Nakashian <ashnakash@gmail.com>
File extensions marked as view (as opposed to edit)
in discovery.xml are now forced to be read-only,
regardless of what the client tries to request.
Change-Id: I3eb00c33ff716800dc317f7377281c6d5f0909d7
Reviewed-on: https://gerrit.libreoffice.org/38480
Reviewed-by: Ashod Nakashian <ashnakash@gmail.com>
Tested-by: Ashod Nakashian <ashnakash@gmail.com>
When saving while editing a cell, the
document is saved with the new cell contents
(as the cell editing is interrupted and applied),
but the modified status is not published to the
client (loleaflet).
This has the unfortunate side-effect that the
EverModified flag is not set in this scenario.
To avoid this, this patch sets the EverModified
upon successful saving. This is easier than
forcing LO Core to emit ModifiedStatus=true
in the scenario above. Though it suffers from
the false-positive result when the user forces
a save. Although this is probably defensible,
since technically the document changes upon
saving (though not the content if there was
nothing new to save). On the other hand,
it's less harmful to signal modification
due to save than to ignore genuine
modification to the document.
Change-Id: I7ccba62731f534b69819f7924eeb9231d874c111
Reviewed-on: https://gerrit.libreoffice.org/38398
Reviewed-by: Ashod Nakashian <ashnakash@gmail.com>
Tested-by: Ashod Nakashian <ashnakash@gmail.com>
