Troubleshooting

Precompilation Issues

  • Dependency Conflicts: After updating the repository, new dependencies may cause precompilation to fail. Ensure all packages are fully resolved by running:
include("resolve.jl")
recurse_projects() # Optionally set update=true
  • Starting the REPL: Rather than starting a REPL and then activating the project, launch Julia directly with the project as an argument to avoid precompilation issues:
julia --project=./Planar

  • Python-Dependent Precompilation: Precompiling code that relies on Python, such as exchange functions, may lead to segmentation faults. To prevent this:

    • Clear global caches, like TICKERS_CACHE100, before precompilation. Ensure global constants are empty, as their contents are serialized during precompilation.
    • Avoid using macros that directly insert Python objects, such as @py, in precompilable functions.

  • Persistent Precompilation Skipping: If a package consistently skips precompilation, check if JULIA_NOPRECOMP environment variable includes dependencies of the package.

  • _debug_ not found happens when trying to precompile a strategy with debug enabled (JULIA_DEBUG="all") while the module (Planar or a submodule like SimMode) has been not precompiled with debug enabled.

Python Module Discovery

  • Missing Python Dependencies: If Python reports missing modules, execute the following in the Julia REPL with the current repository activated:
; find ./ -name .CondaPkg | xargs -I {} rm -r {} # Removes existing Conda environments
using Python # Activates our Python wrapper with CondaPkg environment variable fixes
import Pkg; Pkg.instantiate()
  • Force CondaPkg Environment Resolution: In the case of persistent issues, force resolution of the CondaPkg environment by running:
using Python.PythonCall.C.CondaPkg
CondaPkg.resolve(force=true)

Then, restart the REPL.

Unresponsive Exchange Instance

  • Idle Connection Closure: If an exchange instance remains idle for an extended period, the connection may close. It should time out according to the ccxt exchange timeout. Following a timeout error, the connection will re-establish, and API-dependent functions will resume normal operation.

Data Saving Issues

  • LMDB Size Limitations: When using LMDB with Zarr, the initial database size is set to 64MB by default. To increase the maximum size:
using Data
zi = zinstance()
Data.mapsize!(zi, 1024) # Sets the DB size to 1GB
Data.mapsize!!(zi, 100) # Adds 100MB to the current mapsize (resulting in 1.1GB total)

Increase the mapsize before reaching the limit to continue saving data.

Misaligned Plotting Tooltips

  • Rendering Bugs: If you encounter misaligned tooltips with WGLMakie, switch to GLMakie to resolve rendering issues:
using GLMakie
GLMakie.activate!()

Segfaults when saving ohlcv

The default ZarrInstance uses an LMDB store. It is possible that the underlying lmdb database has been corrupted. To fix this the database must be re-created. Either delete the database manually (default path is in under Data.DATA_PATH) or run this code on a fresh repl:

using Data
Data.zinstance(force=true)

LMDB not available

LMDB requires a precompiled binary. If it is not available for your platform you can disable it by setting the data_store preference in your strategy (or top package) Project.toml.

[preferences.Data]
data_store = "" # Disables lmdb (set it back to "lmdb" to enable lmdb)

Debugging in vscode

To trigger breakpoints inside functions wrapped with a custom logger, like when executing a strategy with start!(::Strategy), it might be necessary to disable loading of Base.CoreLogging compiled module. Inside your user settings modify the option julia.debuggerDefaultCompiled as shown below:

    "julia.debuggerDefaultCompiled": [
        "ALL_MODULES_EXCEPT_MAIN",
        "-Base.CoreLogging"
    ],