On 9/7/22 10:56, Daniel Tschlatscher wrote: > This patch series' aim is to add a download button in the tasklog- > viewer GUI so that users may access all parts of the tasklog more > easily (The tasklog-viewer only displays 50 lines at a time). > > For this change in the GUI (proxmox-widget-toolkit) there were revised > implementations for all backends (PVE, PMG and PBS) needed. > Now, when the parameter 'limit' for the /log API call is set to '0', > the backends will stream the file rather than returning a JSON array > with the corresponding lines. (Beforehand, it would return 0 lines) > > I initially included a number of patches aimed at refactoring > redundant implementations of downloading a file via a button click. > However, after talking with Sterzy, Dominik and Aaron a bit about > it, I decided to only send the "core" task log download patch series > now, because the refactoring depends on the changes made in widget- > toolkit, but is done in the same repositories as the backend patches. > This would make both reviewing and applying them more confusing, and > the refactoring patches are "a step removed" anyway from this patch > series. Therefore, I am going to send a separate patch series for them > later instead. > > > During the implementation of the file stream download I found a bug > concerning chromium browsers because of which downloads "randomly" > fail. > The problem was, when implementing the download through an html a tag > and setting its property "download" to anything, (even undefined) > with a not imported self-signed certificate, the download will fail > once the SSL connection is reset. > In my testing this was the case after about 5 seconds. This problem is > especially pronounced in the PMG as there are less background calls in > contrast to the PVE. The PBS, however, does not seem to suffer this > error at all, because the SSL handshake timeout seems to not run out. > > Without the 'download' attribute you can't set the filename though. > The file will only have a proper name if the server adds the correct > 'content-disposition' header. A few calls to the 'downloadAsFile' > function set the filename anyway, because in most cases the browser > frontend is making enough polling calls so that the SSL handshake > does not have to be executed for the download. (That is also why it > probably did not come up before) > This problem does not apply in Firefox as far as I could tell. > > See https://bugs.chromium.org/p/chromium/issues/detail?id=993362 for > more information on this. > > -- Changes from v1 -- > * Rewrote commit messages and cover letter > * Got rid of some refactoring which were initially included > (going to send as own patch series later) > * Rebased to current repository versions > > > Daniel Tschlatscher (1): > make tasklog downloadable in the backup server backend > > src/api2/node/tasks.rs | 159 ++++++++++++++++++----------- > 4 files changed, 107 insertions(+), 79 deletions(-) > > Daniel Tschlatscher (1): > stream file method for PMG and PVE > > src/PVE/Tools.pm | 28 ++++++++++++++++++++++++++++ > 1 file changed, 28 insertions(+) > > Daniel Tschlatscher (1): > acknowledge content-disposition header > > src/PVE/APIServer/AnyEvent.pm | 3 +++ > 1 file changed, 3 insertions(+) > > Daniel Tschlatscher (1): > Revised task log API call when parameter 'limit' is 0 > > PVE/API2/Tasks.pm | 13 ++++++++++--- > 3 files changed, 15 insertions(+), 24 deletions(-) > > Daniel Tschlatscher (1): > revised task log download function in the PMG backend > > src/PMG/API2/Tasks.pm | 34 +++++++++++----------------------- > 1 file changed, 11 insertions(+), 23 deletions(-) > > Daniel Tschlatscher (3): > Source file download call in central function > add task log download button in TaskViewer > > src/Utils.js | 13 +++++++++++++ > src/window/TaskViewer.js | 17 +++++++++++++++-- > src/window/FileBrowser.js | 11 +++++------ > 3 files changed, 38 insertions(+), 8 deletions(-) >
Other than one bug in `pmg-api` (see my other email) this series seems to work as intended. The bug in `pmg-api` is fixed fairly easily by just calling the right function and I tested that locally already. So unless there are other considerations: Tested-by: Stefan Sterz <s.st...@proxmox.com> _______________________________________________ pve-devel mailing list pve-devel@lists.proxmox.com https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
