Storage and jobs
Glossa separates the working database, downloaded file repository and network cache. A persistent queue handles long operations without requiring their screens to remain open.
Data locations
| Component | Contents | Storage |
|---|---|---|
| Database | Projects, translations, catalogue, resources and settings | Local data directory |
| File repository | Manifests, images, thumbnails and local versions | Selectable directory, including an external drive |
| Network cache | Reusable search responses and images | Temporary storage with a configurable limit |
View and change locations under Settings → Data. Changing the data directory copies the database, checks the copy and records the new location for the next restart; the original is not deleted automatically. Changing the file repository selects an empty directory or reconnects an existing Glossa repository without transferring files from the previous location.
Keep the database on a local drive outside synchronised folders. The file repository can be synchronised only when files remain physically available on disk; cloud placeholders do not guarantee that images can be read.
Job lifecycle
The status-bar panel lists queued, running and finished jobs. Details include progress, the current step, attempts, timestamps and errors. Downloads, image reduction, repository checks and searches have dedicated jobs.
| State or command | Behaviour |
|---|---|
| Waiting | The job is waiting for resources or a remote rate limit to expire |
| Pause | Requests a stop at the next checkpoint |
| Resume | Continues a paused job |
| Cancel | Ends the job; files already saved remain available |
| Retry | Starts another attempt after a failure |
A user-requested pause is distinct from waiting for an automatic retry. A retry countdown is not an estimate of the time required to finish the job.
Closing the application pauses active jobs after confirmation. They remain paused on restart. Settings → Jobs can enable automatic download recovery and configure limits by resource category.
Image dimensions
Settings → Library → Images defines page and thumbnail dimensions and image-reduction settings. A per-work selection overrides the general setting. The numeric value specifies the desired long edge, not guaranteed dimensions for the received files.
| Library mode | Request rule |
|---|---|
| Automatic | Uses the nearest declared halving level when available; otherwise calculates width from the desired long edge |
| Ready-made sizes only | Uses declared halving levels; requests full size if none are available |
| Exact size | Calculates a proportional width from the requested long edge |
| Maximum resolution for the work | Requests full size regardless of the library mode |
Pages smaller than the target and pages without valid dimensions are requested at full size. A predefined size can be larger or smaller than the target. If the service rejects the requested dimensions, a download may use full size and preserve it without local resizing. Thumbnails are generated from downloaded pages.
Requesting a different resolution replaces the downloaded one: for a work Glossa keeps a single image copy at a time (see “One copy per work” further down). Saving a page from the viewer instead uses the image already loaded, without requesting a new one.
Network profiles
Settings → Library → Settings manages concurrency, requests per minute, cooldown after a rejection, retry limits and timeouts. Changes require an explicit save. The Libraries tab assigns a profile and image request mode to each service.
Built-in profiles can be edited but not deleted. A custom profile must be unassigned before deletion. Concurrent requests to an individual host are capped at four, with capacity reserved for interactive reading. Job limits do not override these network constraints. The status-bar indicator shows active and waiting requests and image provenance.
Coverage and integrity
Local page counts are derived from repository files. The expected total comes from the manifest and is recorded when the viewer opens as well as during downloads. Pages reported as unavailable by the service are counted separately; completeness refers to pages that can be retrieved.
A quick check verifies file presence. A full check reads file contents, validates their format and compares available checksums. Both report intact, missing, corrupt and unassociated files. Neither automatically deletes or downloads files. Cleaning up files without an associated work is a separate command that rechecks the repository before proceeding.
Image reduction and caching
Recompression rewrites the copy's pages in place, at a lower quality and the same dimensions: same pixels, fewer bytes, no second copy of the book. It is not reversible — getting the original quality back means downloading again from the library (details further down). If some pages cannot be processed, the job reports an error and retains the pages produced successfully.
The network cache reuses responses and images. Its default size limit is 512 MB, and search responses are valid for 24 hours by default. Images are subject to the size limit without the same time-based expiry. Change these values or clear the cache under Settings → Data. Cached data does not increase downloaded-page counts and is excluded from backups.
One copy per work
A work keeps one image copy, at the size chosen when downloading, with one file per page. Asking for the book at another size says so first and replaces what you have: the old sizes are deleted only once the download succeeds, so a network failure does not leave you with nothing.
The repository can still hold several sizes at once, but no command creates more than one.
Waiting on libraries
Every request to a library waits for its turn: that is what stops Glossa from knocking too fast and getting refused. Every wait has a deadline: twenty seconds for what you are looking at, eight for background checks. Once it expires the request gives up and says so — “not verified” — instead of hanging.
This addresses a concrete case: after a refusal, some libraries ask you to wait for minutes. Without a deadline a single request started at that moment stayed in the queue for all that time, and everything else concerning that library — including adding works — queued up behind it.
The document next to the images
When the library serves the work as a single document, that file lives in the repository next to the image pages of the same work, not among the resolutions: it is another copy, not another resolution. It takes space of its own, is deleted on its own, and does not disappear when you free the images.
The download is a job like the others and respects the same library network limits: the file is written to a staging area, checked — signature and proper ending — and only then enters the repository, so a dropped connection never leaves half a document among your files. Pages are counted from the document as soon as it arrives. If the library declared a different number, the count from the file wins, and the difference stays in the operations log without on-screen warnings.
Repository verification checks the document the way it checks pages, comparing the checksum recorded when it arrived.
Acting on the page you are reading
The commands for a single page live in the right-hand panel, under Digital copies, in the section at the top about the page open in the viewer. They are icon commands: the name appears on hover. They stay visible, disabled, when the viewer is showing another copy.
With a page open you can:
- download it, even when the book is not on disk: the page goes into the folder of the resolution chosen for that work;
- download it at maximum resolution: it is requested again and replaces the one present, staying the only file for that page;
- download it again at the book resolution, which recovers the space when the detail is no longer needed;
- delete it from disk: the page is excluded and does not come back with a new download of the book, nor on its own. Asking for it again readmits it.
When the book is already at maximum resolution both resolution commands are disabled: they would request the same image.
Below, the real size of that page — the pixels it actually has, which after a retake are no longer the book's — and its weight. The viewer toolbar keeps only the reading commands: local-only reading, zoom, thumbnails and the link to the page on the library site.
The commands about the pages on disk — check, recompress, delete — sit on the resolution row, not in the section header: that is where you can see what they act on.
On the item page the copy states how many pages you removed on purpose: without that line an incomplete copy would look broken.
Making the pages lighter
The recompression command rewrites every page of the copy at a lower quality, without changing its dimensions: it is for when the size is right and space is the problem. It does not create a second copy of the book and it is not reversible — to get the previous quality back you download from the library again.
If a page turns out to need better quality after that, take it at the highest resolution: it replaces the recompressed one.
Messages and system log
The bottom panel holds three tabs: messages from the running translation, the system log, and jobs. The panel command in the status bar always opens it on the system log, which is available in every area and shows what the program wrote while working — library searches, downloads, storage, saves — reading the application log file directly, including the rotated files from previous sessions.
Filters narrow by area (Library, Translation, Jobs, Interface) and by level (error, warning, info, debug); the search box works on the text of each line. Lines produced by third-party libraries — database queries, keyring, network connections — stay hidden until explicitly requested: on their own they are most of the file.
New lines arrive by themselves while the tab is open, at the top of the list; whoever is reading further down does not see them shift under their eyes. The data of a line — the part in curly braces — is colored as in a code editor: field names, strings, numbers and values each in their own color. The braces command turns the coloring on and off, and the choice is remembered.
"Clear the view" hides the lines currently shown without touching the file on disk: lines arriving afterwards keep appearing, and reloading brings them all back. "Load earlier messages" continues reading backwards. The log folder path is shown in the in-app guide, under troubleshooting.
Job history and retention
The bottom panel is the operational view: it shows jobs that have not finished yet plus those completed in the last 24 hours, with pause, resume and retry. The bin command in its header takes the finished rows out of the view, so new jobs read clearly: it deletes nothing, and a hidden job that starts again comes back on its own. The cleanup lasts for the current session.
The full history lives in the right-hand column of the Overview: it lists everything that went through the queue, reads in pages and can be narrowed, widened or closed like the other side columns. Every row expands like in the bottom panel, with phase, attempts, times, outcome and error.
At the top of the column there is a free-text search over the job name and two rows of commands: the first filters by outcome (running, succeeded, failed, interrupted), the second by kind of work. Several choices can be active at once; the state of each row is a symbol explained on hover, not a word.
No job is ever deleted automatically and there is no cap on the number of rows kept. Deletion is always explicit: a single row, or — with the bin at the top of the column — every finished job the filters are showing, so all of them when the filters are off and only the selected ones otherwise. Jobs another surface still depends on stay — today the executions of a stored search, which will be removed together with the search itself.