Update v2.4.0 kh22 rf #2
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
avoid taking a lock for time lookup, just use the time acquired below
fix mostly compiler warnings, the log level check was incorrect although not an error in valid case
a cleanup was not right with the global http-headers. The previous routine specified referred to the member but the replacement missed out the reference.
any client that already has a respcode is not taking any conditional action based on the response from the auth server.
use the aux_data member instead of the refbuf to pass the control structure for intro content building in listener_add. Functionally the same, it just avoids the overlay aspect. uses something more temporary. Actually set a default prefix for processing client headers beforehand
allow for a empty name to used for appending on the end of the items. Needs extra work to be complete though
rare setup but can occur. make auth create the lock and set the count as a minimum in initialising part as the auth_release expects it
To have a cleaner exit path for request going via curl handles, use the callback routines (used for progress reports) to check the status of the servers running state. If it is shutting down then do not immediately abort those transfers, but do after a number of seconds have elapsed. While an immediate abort would stop quicker with in-flight requests, by allowing a certain duration to elapse means allowances for tracking state exterally is kept.
This pushes the #ifdef into the module for the get auth part.
first is to use the aux_data to store the useagent for the mount_add/remove triggers instead of the shared_data. There is no parser in those cases so you cannot use that, and it was used pre-aux_data so use that as it is a better fit for now. Could become a struct if more needed later. url auth was the only auth to utilise this. We used aux_data for holding the name for the command but would leak in certain cases. The command is copied because it could of been aliased and a delayed auth needs to reference it later in the auth thread. In failure cases the default send routines no nothing about the content of aux_data so cannot touch it so would leak. This instead pushes a reserved taken into parser __admin_cnd and uses that for storage which is cleaned up automatically during close.
The http part is essentially the same, the difference is in relation to dividers and end of pariing characters. The http complete also explicitly adds a blank line because the complete routine does not add the end of paring characters to the last entry. This waym for http you get the blank with nothing following it but the previous has a \r\n, in the POST case there is nothing added on the end so no & is at the end. The POST variants don't refer to a client refbuf, instead it returns one to allow the caller to decide what to do. auth url does not use a client for example. YP does not either.
The http work allocates the block at the complete stage and inserts at the head of the list for the client. This means that the original requirments for block processing at the start has changed. Before we had a block ready to go when entering the request routines but now that is now dropped. One to watch though is on incoming streams from sources. They can send stream data after the connetion so you need to keep that around to avoid any misalignment problems.
avoid a sync issue with client flags and make sure the worker is woken up so cleanup can occur.
a chunk of this is the conversion over to using the http work, which builds the post fields which are passed into the libcurl handle. The callback routines for headers and data were not that great. Most of it stems from the problem that headers are not required to be nul-terminated, but it's not clear on whether the buffer passed is really changeable. So to be clearer on this, and to make processing more in-line with strings with nuls, these routines were re-worked to do the same sort of thing be to make sure there is not gotchas
when the relay is established, set the recheck hosts initial time else it will retry immediately. Most likely not something of concern just some extra processing work at the beginning but the feed may sound like a small burst of audio at switchover if the higher priority feed just drops for a moment.
allow a larger range of relays to start up. not excessive but allows more in case many are configured. introduce a start member for relay so that we can process a the startup routine without kicking off a relay early. This is targetted at relays that are offline but we may need to do internal housekeeping. not used much for now.
This is to specify the number of seconds a terminating stream stays active for before it finally shuts down. This has a few possibilities for later commits but if a relay terminates, you may want to delay the reconnection by a short time to allow the sources to reconnect upstream. The basic idea is the the stream put into a state where no further stream reading is performed but all other actions are. The main change in this commit is the splitting of the source_read routine into 2, one being a wrapper for catching the returns from the internal one The linger facility will also be able to interact with the switchover mechanism.
Next update on the routines. The mechanisms for switchover follow a pattern of set the SOURCE_SWITCHOVER and set aux_data to refer to the new client then wait for the original client to swap it over. The odd routine here is the relay_reactivated routine which if it finds a relay for the source that is terminating then there will be no relay client. In this case, we set can set source client as there is no race between us, the listeners and the another client. The switchover routines for the relay client alway proceed to client free up but it is just a matter of whether the relay is freed (priority host switchover) or not (eg source override of relay). With the linger-for setting, we can now have a source linger for a time after termination and subsequent connection by a source client that subsequent connection is treated as a switchover status and jumps in meaning another mechanism for changing sources without a fallback The previous update for switchover did not handle inactive on-demand relays so now a source can override an on-demand relay and drop back when finished The linger-for setting does apply even in a multi-host relay. This is something that may need addressing later.
This may be affecting the safari connections.
here we keep a start and end position in the connection_t and use that to cause a stream stop. we take into account the difference on files and streams, and if there is anything to skip over at the beginning of a file. The oddball here is safari with its multiple requests to assess things, but hopefully that can be tightened up to sort it.
In cases with many quickly rescheduling clients, the limiter was far too high to trigger a reset.
While the connection error is often used to mark the end of a stream in a few cases we may want to stop the current stream and restart another on the same connection, these markers allow for that.
It still leaves the infamous double bandwidth stream bug, but should play at least
a lock was missing on a file handle. This could lead to an abort or hang depending on the underlying locking implementation. some small re-arrangement on initial byte assessment, not a significant issue but is cleaner.
traditionally, this is generated by icecast based on the specified hostname and other settings. While this works well, some may want to be explicit in what is provided to any directories Note that directories will still do checks so if the provided address does not match the incoming request then it can be rejected, subject to any policies the directory uses
…1. Fix identified in karlheyes#422 karl
safari is a problem child really, but use a redirect trick to add a query param to mitigate caching effects. Try out forcing any aacp content type to aac but still allow the _hdr setting to override. karl
This was out of sight for some time but the main components were correct. The memory block allocated has been made slightly larger but now can be expanded on the rare cases there is a need.
…arlheyes#242 Use the css aspects to use the in-built browser use which allows us to avoid the extra requests to get the images. The other part of the pull is the multi-user shoutcast compatible handling. That already exists using a similar type of layout but whether there is something different in the implementation will need to assessed. The web page changes though are fine. karl.
Some metrics in the xls we imported were hidden: - icestats/sources - icestats/clients - icestats/listeners This choice wasn't ours: we just imported a remote xsl file from the Internet. Today we want to be able to graph these metrics to we unhide them here.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.